ÿ‚ �'Nœ 8@p€àÁ !‡ /hYW”i!1/}!1 /}!1ͽ!1œͽ!11Ã!1¤1Ã!1gÆ!1gÆ!1ñÏ!1œñÏ!1'Ó!1h'Ó!1ç!1ôç!1Ÿ'"1pŸ'"1-"1x-"190"1ä90"1Ã9"1pÃ9"1ù<"1<ù<"1ÓP"1ÈÓP"1q‘"1D q‘"1Õ–"1L Õ–"1 š"1¸  š"1•£"1D •£"1˦"1 ˦"1 Ô"1œ  Ô"1#Ö"1x#Ö"1±Ø"1±Ø"1ý"1‚ý"1S*#1ÚS*#1‰-#1ª‰-#1m7#1Ú m7#1)l#1ê )l#1µˆ#1Jµˆ#1��#1��#1ÍŽ#1JÍŽ#1Ë™#1zË™#1½$1ª½$1•ò$1J•ò$1ÏÞ%1jÏÞ%1Iì%1BIì%1ó€&1ró€&1ÃR'1 ÃR'1ke'12"€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P�i b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¥å bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P�i b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¥å  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P�i  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¥å bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes1152patternContiguousinvocations1€Ê bJoperation.3exchange�½\"R$º`bª@}ª¸õŠ0 ò4 i š¤ ŠØ Š š@štЍŠÜššD’’<¢p¢¤’Ø’ ¢@¢t’¨’Ü¢¢D ’x!‚›"¾#á$’%"&&²H'Bk(Ò�)b°*òÒ+‚õ,-¢:.2]/Â0R¢1âÄ2B3Ò)4bL5òn6‚‘7´8¢Ö92ù:Ò;b><a=¢ƒ>2¦?ÒÈ@bëA+BjgC §DrãEÊFr[GÊ—HrÓI JCK²~L²MÂæNÒOÒNPâ‚Qâ¶RòêSòTSU‡V»WïX"#Y"WZ2‹[2¿\Bó]B'^R[_R�`bÃaRæbâ cÒ+ dbN eòp f‚“ g¶ h iš:j*]kºlJ¢mÚÄnjçoú pŠ,qOrªqs:”tʶuZÙvêûwzx Ayšcz:†{ʨ|jË} î~š:3€ÊU�j•‚Âуb„ºM…ІªÅ‡ˆš=‰2yŠ2­‹ÊèŒÊ�ÊPŽÊ„�ʸ�Êì‘Ê ’ÊT“ʈ”ʼ•Êð–Ê$—ÊX˜ÊŒ™ÊÀšÊô›Ê( œÊ\ �Ê� žÊÄ ŸÊø  Ê, ¡ºO ¢Jr £:• ¤Ê· ¥ZÚ ¦êü §z ¨ B ©šd ª*‡ «º© ¬JÌ ­Úî ®j ¯ú3 °ŠV ±y ²ª› ³:¾ ´Êà µZ ¶ê% ·zH ¸ k ¹š� º*° »ºÒ ¼Jõ ½ê¾z:¿]ÀªÁJ¿¢ûÃB;ÄšwÅò³ÆŠïÇâ+ÈzgÉ£Ê×˪̪FͪzΪ®ϪâЪѪJÒª~Óª²ÔªæÕªÖªN×ÊØÊ4ÙÊhÚÊœÛÊÐÜÊÝÊ8ÞÊlßÊ àÊÔáº÷âJã:=äÊ_åZ‚æê¤çzÇè êéš ê*/ëºQìJtíÚ–îj¹ïúÛðŠþñ!òªCó:fôʈõZ«öêÍ÷zðø ùš5ú*XûºzüJ�ýê¿þzâÿª'Jg¢£Bãšò[J˜¢Ôú úD úx ú¬ úà ú úH ú| ú° úä ú úL ú€ ú´ úè ú úP ú„ ú¸ úì ú úT úˆ ú¼  úð !ê "z6 #jY $ú{ %Šž &Á 'ªã (:)Ê(*ZK+êm,z�- ³.šÕ/*ø0º1J=2Ú_3j‚4ú¤5ŠÇ6ê7ª 8:/9ÊQ:Zt;ê–<z¹=Ü>ªþ?J!@ÚCAzƒBÒ¿CrÿDÊ;E"xFz´GÒðH*-I*aJ*•K*ÉL*ýM*1N*eO*™P*ÍQ*R*5S*iT*�U*ÑV*W*9X*mY*¡Z*Õ[* \*=]*q^*¥_*Ù`* a0bªRcšud*˜eººfJÝgÚÿhj"iúDjŠgkŠlª¬m:ÏnÊñoZpê6qzYr |sšžt*ÁuºãvJwÚ(xjKyúmzŠ�{³|ªÕ}Jø~Úz=€ `�ªŸ‚܃¢„úW…R”†ªÐ‡ ˆZI‰Z}ŠZ±‹ZåŒZ�ZMŽZ��Zµ�Zé‘Z ’ZQ “Z… ”Z¹ •Zí –Z!!—ZU!˜Z‰!™Z½!šZñ!›Z%"œZY"�Z�"žZÁ"ŸZõ" Z)#¡JL#¢Ún#£Ê‘#¤Z´#¥êÖ#¦zù#§ $¨š>$©*a$ªºƒ$«J¦$¬ÚÈ$­jë$®ú %¯Š0%°S%±ªu%²:˜%³ʺ%´ZÝ%µêÿ%¶z"&· E&¸šg&¹*Š&ºº¬&»JÏ&¼Úñ&½² ¾B3 ¿ÒU Àbx Á¸ ÂZô Ãú3 ÄRp Ū¬ Æé ÇZ% Ȳa ɲ• Ê²É Ë²ý ̲1 Ͳe β™ Ï²Í Ð² Ѳ5 Ò²i Ó²� Ô²Ñ ÕÊ ÖÊ: ×Ên ØÊ¢ ÙÊÖ ÚÊ  ÛÊ> ÜÊr Ýʦ ÞÊÚ ßÊ àÊB áºe âJˆ ã:« äÊÍ åZð æê çz5 è X éšz ê*� 뺿 ìJâ íÚ îj' ïúI ðŠl ñ� òª± ó:Ô ôÊö õZ öê; ÷z^ ø � ùš£ ú*Æ ûºè üJ ýÚ- þjP ÿúr Š• *Õ ‚ "Q z� ÒÉ * ‚B Ú~ Ú² Úæ Ú ÚN Ú‚ Ú¶ Úê Ú ÚR Ú† Úº Úî Ú" ÚV ÚŠ Ú¾ Úò Ú& ÚZ ÚŽ Ú Úö Ú* Ú^ !Ê� "Z¤ #JÇ $Úé %j  &ú. 'ŠQ (t )ª– *:¹ +ÊÛ ,Zþ -ê  .zC / f 0šˆ 1*« 2ºÍ 3Jð 4Ú 5j5 6úW 7Šz 8� 9ª¿ ::â ;Ê <Z' =êI >Šl ?� @º± AZñ B²- CRm Dª© Eæ FZ" G²^ H › I Ï J  K 7 L k M Ÿ N Ó O  P ; Q o R £ S × T  U ? V s W § X Û Y  Z C [ w \ « ] ß ^  _ G ` { aú� bŠÀ czã d  eš( f*K gºm hJ� iÚ² jjÕ kú÷ lŠ m= nª_ o:‚ pʤ qZÇ rêé sz t / ušQ v*t wº– xJ¹ yÚÛ zjþ {ú ! |ŠC! }f! ~ºˆ! J«! €êÍ! �Š " ‚âI" ƒ‚‰" „ÚÅ" …2# †Š># ‡âz# ˆ:·# ‰:ë# Š:$ ‹:S$ Œ:‡$ �:»$ Ž:ï$ �:#% �:W% ‘:‹% ’:¿% “:ó% ”:'& •:[& –:�& —:Ã& ˜:÷& ™:+' š:_' ›:“' œ:Ç' �:û' ž:/( Ÿ:c(  :—( ¡*º( ¢ºÜ( £ªÿ( ¤:") ¥ÊD) ¦Zg) §ê‰) ¨z¬) © Ï) ªšñ) «** ¬º6* ­JY* ®Ú{* ¯jž* °úÀ* ±Šã* ²+ ³ª(+ ´:K+ µÊm+ ¶Z�+ ·ê²+ ¸zÕ+ ¹ ø+ ºš, »*=, ¼º_, ½J‚, ¾ê¤, ¿zÇ, Àê, Áº)- Âf- ò¥- Ä â- Åb. ƺZ. Ç—. ÈjÓ. Éj/ Êj;/ Ëjo/ Ìj£/ Íj×/ Îj 0 Ïj?0 Ðjs0 Ñj§0 ÒjÛ0 Ój1 ÔjC1 Õjw1 Öj«1 ×jß1 Øj2 ÙjG2 Új{2 Ûj¯2 Üjã2 Ýj3 ÞjK3 ßj3 àj³3 áZÖ3 âêø3 ãÚ4 äj>4 åú`4 抃4 ç¦4 èªÈ4 é:ë4 êÊ 5 ëZ05 ìêR5 ízu5 î ˜5 5 ð*Ý5 ñºÿ5 òJ"6 óÚD6 ôjg6 õú‰6 öЬ6 ÷Ï6 øªñ6 ù:7 úÊ67 ûZY7 üê{7 ýzž7 þÁ7 ÿªã7 J8 êE8 B‚8 âÁ8 :þ8 ’:9 êv9 B³9 šï9 š#: šW: š‹: š¿: šó: š'; š[; š�; šÃ; š÷; š+< š_< š“< šÇ< šû< š/= šc= š—= šË= šÿ= š3> šg> š›> šÏ> !Šò> "? # 8? $šZ? %*}? &ºŸ? 'JÂ? (Úä? )j@ *ú)@ +ŠL@ ,o@ -ª‘@ .:´@ /ÊÖ@ 0Zù@ 1êA 2z>A 3 aA 4šƒA 5*¦A 6ºÈA 7JëA 8Ú B 9j0B :úRB ;ŠuB <˜B =ªºB >JÝB ?ÚÿB @z"C AbC BržC CÞC DRD EªYD F–D GZÒD H²E I²BE J²vE K²ªE L²ÞE M²F N²FF O²zF P²®F Q²âF R²G S²JG T²~G U²²G V²æG W²H X²NH Y²‚H Z²¶H [²êH \²I ]²RI ^²†I _²ºI `²îI a¢J b24J c"WJ d²yJ eBœJ fÒ¾J gbáJ hòK i‚&K jIK k¢kK l2ŽK m°K nRÓK oâõK prL q;L r’]L s"€L t²¢L uBÅL vÒçL wb M xò,M y‚OM zrM {¢”M |2·M }ÂÙM ~büM š €:: �Úy ‚2¶ ƒÒõ „*2 …‚n †Úª ‡2ç ˆŠ# ‰ŠW ŠŠ‹ ‹Š¿ ŒŠó �Š' ŽŠ[ �Š� �ŠÃ ‘Š÷ ’Š+ “Š_ ”Š“ •ŠÇ –Šû —Š/ ˜Šc ™Š— šŠË ›Šÿ œŠ3 �Šg žŠ› ŸŠÏ  Š ¡z& ¢ I £úk ¤ŠŽ ¥± ¦ªÓ §:ö ¨Ê ©Z; ªê] «z€ ¬ £ ­šÅ ®*è ¯º °J- ±ÚO ²jr ³ú” ´Š· µÚ ¶ªü ·: ¸ÊA ¹Zd ºê† »z© ¼ Ì ½ªî ¾J ¿ê3 ÀŠV Á*– Â‚Ò Ã" ÄzN ÅÒŠ Æ*Ç Ç‚ ÈÚ? ÉÚs ÊÚ§ ËÚÛ ÌÚ ÍÚC ÎÚw ÏÚ« ÐÚß ÑÚ ÒÚG ÓÚ{ ÔÚ¯ ÕÚã ÖÚ ×ÚK ØÚ ÙÚ³ ÚÚç ÛÚ ÜÚO ÝÚƒ ÞÚ· ßÚë àÚ áÊB âZe ãJˆ äÚª åjÍ æúï çŠ è5 éªW ê:z ëÊœ ìZ¿ íêá îz ï ' ðšI ñ*l òºŽ óJ± ôÚÓ õjö öú ÷Š; ø^ ùª€ ú:£ ûÊÅ üZè ýú  þš- ÿ:P Úr z² Òî r. Êj "§ zã Ò *\ *� *Ä *ø *, *` *” *È *ü *0 *d *˜ *Ì * *4 *h *œ *Ð * *8 *l *  *Ô * *< !_ "ª� #š¤ $*Ç %ºé &J  'Ú. (jQ )ús *Š– +¹ ,ªÛ -:þ .Ê  /ZC 0êe 1zˆ 2 « 3šÍ 4*ð 5º 6J5 7ÚW 8jz 9úœ :Š¿ ;â <ª! =J'! >ÚI! ?zl! @ �! AªÎ! B " C¢J" Dú†" ERÃ" Fªÿ" G<# HZx# IZ¬# JZà# KZ$ LZH$ MZ|$ NZ°$ OZä$ PZ% QZL% RZ€% SZ´% TZè% UZ& VZP& WZ„& XZ¸& YZì& ZZ ' [ZT' \Zˆ' ]Z¼' ^Zð' _Z$( `ZX( aJ{( bÚ�( cÊÀ( dZã( eê) fz() g K) hšm) i*�) jº²) kJÕ) lÚ÷) mj* nú<* oŠ_* p‚* qª¤* r:Ç* sÊé* tZ + uê.+ vzQ+ w t+ xš–+ y*¹+ zºÛ+ {Jþ+ |Ú , }zC, ~ f, ªˆ, €:«, �Úê, ‚2'- ƒÒf- „*£- …‚ß- †Ú. ‡2X. ˆŠ”. ‰ŠÈ. ŠŠü. ‹Š0/ ŒŠd/ �Š˜/ ŽŠÌ/ �Š0 �Š40 ‘Šh0 ’Šœ0 “ŠÐ0 ”Š1 •Š81 –Šl1 —Š 1 ˜ŠÔ1 ™Š2 šŠ<2 ›Šp2 œФ2 �ŠØ2 žŠ 3 ŸŠ@3  Št3 ¡z—3 ¢ º3 £úÜ3 ¤Šÿ3 ¥"4 ¦ªD4 §:g4 ¨ʉ4 ©Z¬4 ªêÎ4 «zñ4 ¬ 5 ­š65 ®*Y5 ¯º{5 °Jž5 ±ÚÀ5 ²jã5 ³ú6 ´Š(6 µK6 ¶ªm6 ·:�6 ¸ʲ6 ¹ZÕ6 ºê÷6 »z7 ¼ =7 ½ª_7 ¾:‚7 ¿Ú¤7 ÀjÇ7 Á 8 ÂbC8 Ã8 ÄZ¿8 Ųû8 Æ 89 Çbt9 Ⱥ°9 ɺä9 ʺ: ˺L: ̺€: ͺ´: κè: Ϻ; кP; Ѻ„; Òº¸; Óºì; Ôº < ÕºT< Öºˆ< ׺¼< غð< Ùº$= ÚºX= ÛºŒ= ܺÀ= ݺô= Þº(> ߺ\> àº�> ᪳> â:Ö> ã*ù> äº? åJ>? æÚ`? çjƒ? èú¥? éŠÈ? êë? ëª @ ì:0@ íÊR@ îZu@ ïê—@ ðzº@ ñ Ý@ òšÿ@ ó*"A ôºDA õJgA öÚ‰A ÷j¬A øúÎA ùŠñA úB ûª6B ü:YB ýÚ{B þjžB ÿ ÁB šãB :#C ’_C 2ŸC ŠÛC âD :TD ’�D êÌD êE ê4E êhE êœE êÐE êF ê8F êlF ê F êÔF êG êL 4ê`L 5zƒL 6 ¦L 7šÈL 8*ëL 9º M :J0M ;ÚRM <juM = ˜M >šºM ?:ÝM @ÊÿM Az?N BÒ{N C‚»N DÚ÷N EB4O FšpO G­O HZéO IjP JjQP Kz…P Lz¹P MŠíP NŠ!Q OšUQ Pš‰Q Qª½Q RªñQ Sº%R TºYR UÊ�R VÊÁR WÚõR XÚ)S Yê]S Zê‘S [úÅS \úùS ] .T ^ bT _–T `ÊT a íT bšU cŠ2U dUU eªwU f:šU gʼU hZßU iêV jz$V k GV lšiV m*ŒV nº®V oJÑV pÚóV qjW rú8W sŠ[W t~W uª W v:ÃW wÊåW xZX yê*X zzMX { pX |š’X }:µX ~Ê×X júX €úY �ª\Y ‚™Y ƒ’ØY „êZ …RQZ †ª�Z ‡òÉZ ˆJ[ ‰Z:[ ŠZn[ ‹J¢[ ŒJÖ[ �Z \ ŽZ>\ �Jr\ �J¦\ ‘ZÚ\ ’Z] “JB] ”Jv] •Zª] –ZÞ] —J^ ˜JF^ ™Zz^ šZ®^ ›Jâ^ œJ_ �ZJ_ žZ~_ ŸJ²_  Jæ_ ¡: ` ¢Ê+` £ºN` ¤Jq` ¥Ú“` ¦j¶` §úØ` ¨Šû` ©a ªª@a «:ca ¬Ê…a ­Z¨a ®êÊa ¯zía ° b ±š2b ²*Ub ³ºwb ´Jšb µÚ¼b ¶jßb ·úc ¸Š$c ¹Gc ºªic »:Œc ¼Ê®c ½ZÑc ¾êóc ¿zd 7 :z2breasonOperatorInputsvalue7invocations1 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1 bšWuoperation.4ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue8bytes2304patternContiguousinvocations1€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 5uªê_ bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1<5µêjŸTÔ ‰¾>só(¨Ý]’GÇü|±ñ& f › Û  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)")B)Z-b1reasonLocalCopyvalue10bytes512patternContiguousinvocations4 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)")B)Z-b1reasonLocalCopyvalue10bytes256patternContiguousinvocations1  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10FF Õ ÕLLÍÍ)D*D4»4»>2?2I§I§ST^�^�his‡s‡}" b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 # b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1!$ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5" €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðFF½½44«« " " ± ±((ŸŸ��{{òòi i"à"à$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbìbìdaeagÖgÖiKjKlÀlÀnAoAq¶q¶s+t+v v xy{Š{Š}ÿ}ÿt€t‚é‚é„^…^‡Ó‡Ó‰HŠHŒ½Œ½Ž2�2‘§‘§“#* b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1$+ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxi!1 ~!1è ~!1«¾!1d«¾!1MÅ!1lMÅ!1×Î!1è×Î!1 Ò!1´ Ò!1ßç!1@ßç!1}("1¼}("1/"1Ä/"1©8"1@©8"1ß;"1 ß;"1±Q"1˜±Q"1O’"1O’"1ñ˜"1 ñ˜"1{¢"1˜ {¢"1±¥"1d ±¥"1óÒ"1ð óÒ"1[Ú"1Ì [Ú"1»þ"1P »þ"1ý+#1pý+#1…:#1L…:#1Ao#1ØAo#13�#1ü3�#11š#1(1š#1±$1´±$1‰ò$1x‰ò$1ÃÞ%1|ÃÞ%1=ì%1\=ì%1½€&1è½€&1�R'1,�R'15e'10€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¡ÕvB b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I‰µõ¾þ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¡ÕvB b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I‰µõ¾þ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¡ÕvB  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I‰µõ¾þ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1¸´/Oì|´h bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1%e]�ú:™Ù8x  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d%¥]Ýà`:º›—VÖôtÑQ®.‹ ý=hèGÇ&¦ … ä d à C ¢ " �  ` à =}Úy¹X bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛPPÇÇ)<*<4™4™>'?'I›I›ST^�^�hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››……úúo o"ä"ä$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=Õ=Õ?J@JB¿B¿D4E4G©G©I7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`KbËbËd?e?g³g³i'j'l›l›noqƒqƒs÷s÷ukvkxßxßzS{S}Ç}Ç;€;‚¯‚¯„#…#‡™‡™‰ŠŒƒŒƒŽøŽø�m‘m“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„i!1/}!1/}!1ͽ!1|ͽ!11Ã!1„1Ã!1gÆ!1ðgÆ!1ñÏ!1|ñÏ!1'Ó!1H'Ó!1ûæ!1Ôûæ!1™'"1P™'"1-"1X-"190"1Ä90"1Ã9"1PÃ9"1ù<"1ù<"1ÍP"1¨ÍP"1k‘"1$ k‘"1Õ–"1, Õ–"1 š"1˜  š"1•£"1$ •£"1˦"1ð ˦"1 Ô"1|  Ô"1‘Ú"1X‘Ú"1ñþ"1Ôñþ"13,#1ô3,#1%:#1Ð%:#1án#1\án#1Y�#1€Y�#1W˜#1¬W˜#1«$18«$1ƒò$1üƒò$1½Þ%1½Þ%17ì%1à7ì%1�€&1l�€&1]R'1Ì]R'1e'1Ѐ} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ošf b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¥å bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ošf b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¥å  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ošf  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¥å bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ­Í¥Å€  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1G‡]ÎS“Ø  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dGÇ�ÒRއÕU<¼�dä+«òr¹9~þCÈ™ÙÏO ”  Y Ù  ž ã c ª ê 1 q ¸ ø =} bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10FF Õ ÕKKÈÈ)=*=4²4²>'?'IšIšS T ^Œ^Œhÿhÿrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7operation.6exchange :z2breasonOperatorInputsvalue8invocations1# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%w©operation.6ipu_stack_attention_merge_large_query_f165:5z9299ZÒ9z="=zA"AzE*EbIreasonOperatorKernelvalue10kernelSpecoutputElements576input0Elements640input1Elements6272invocations1ÇðFF¼¼22¨¨   ­ ­##™™……ûûqqçç] ]"Ó"Ó$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I+J+LžLžNOQ„Q„S÷S÷UjVjXÝXÝZP[P]Ã]Ã_6`6b©b©deg�g�ijlulunônôpgqgsÚsÚuMvMxÀxÀz3{3}¦}¦€‚Œ‚Œ„ÿ„ÿ†r‡r‰å‰å‹XŒXŽËŽË�>‘>“2Planned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })çxi!1~!1è~!1±¾!1d±¾!1MÅ!1lMÅ!1×Î!1è×Î!1 Ò!1´ Ò!1ßç!1@ßç!1}("1¼}("1/"1Ä/"1©8"1@©8"1ß;"1 ß;"1±Q"1˜±Q"1O’"1O’"1ñ˜"1 ñ˜"1{¢"1˜ {¢"1±¥"1d ±¥"1ïÓ"1ð ïÓ"1Ú"1È Ú"1yþ"1D yþ"1·,#1d·,#1©:#1<©:#1eo#1Èeo#1ç�#1ìç�#1åš#1åš#1¥$1¤¥$1}ò$1h}ò$1·Þ%1l·Þ%11ì%1L1ì%1u€&1Øu€&1ER'1ER'1íd'1 €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖxD b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Jжö¿ÿ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖxD b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Jжö¿ÿ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖxD  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Jжö¿ÿ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ˜” Í bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1h¨ äP¶ö\œ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dhèPÐ8¸  ‘ˆÌL‡ðpBÂØXx¸À@¦&Œ r ò X Ø > ¾ $ ¤  Š ð pØ€À&fÌ  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×LLÅÅ):*:4–4–>$?$I—I—ST^‡^‡hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡üüqqææ[ ["Ð"Ð$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Â=Â?7@7B¬B¬D!E!G–G–I$J$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b­b­deg‘g‘ijlulunçnçpYqYsËsËu=v=x¯x¯z!{!}“}“€‚w‚w„é„é†]‡]‰ЉЋCŒC޶޶�)‘)“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxi!1S}!1èS}!1ñ½!1dñ½!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!1%ç!1@%ç!1Ã'"1¼Ã'"1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1÷P"1˜÷P"1•‘"1•‘"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"1OÚ"1Ì OÚ"1¯þ"1H ¯þ"1ñ+#1hñ+#1�:#1D�:#1Yo#1ÐYo#1‡�#1ô‡�#1…š#1 …š#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1Ï€&1àÏ€&1ŸR'1$ŸR'1Ge'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations15i¶‚ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 D„5uÈ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations15i¶‚ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 D„5uÈ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations15i¶‚  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 D„5uÈ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 †¦"B<\ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Va¡B¥åHˆ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d–aáÍMB„#£;»„åeÆF§'ˆkëNÎ1 ±  ” ÷ w † Æ Ú Z ½ =   �Á"bÅh¨ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ü ÜRRÈÈ)>*>4œ4œ>(?(IœIœST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÜÜVVÐÐ J J Ä Ä>>¸¸22¬¬""˜˜ „ „"ú"ú$p%p'æ'æ)\*\,Ò,Ò.H/H1¾1¾34446ª6ª8 9 ;–;–=ô=ô?h@hBÜBÜDPEPGÄGÄIPJPLÄLÄN8O8Q¬Q¬S T V”V”XY[|[|]ð]ð_d`dbâbâdVeVgÊgÊi>j>l²l²n&o&qšqšstv‚v‚xöxözj{j}Þ}ÞR€R‚ƂƄ:…:‡®‡®‰!Š!Œ”Œ”Ž�‘z‘z“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxi!1~!1è~!1¥¾!1d¥¾!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!1Ùç!1@Ùç!1w("1¼w("1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1«Q"1˜«Q"1I’"1I’"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"1gÚ"1Ì gÚ"1Çþ"1P Çþ"1 ,#1p ,#1C4#1LC4#1íw#1Èíw#1«†#1쫆#1©‘#1È©‘#1“$1T“$1kò$1kò$1¥Þ%1¥Þ%1ì%1üì%1½€&1ˆ½€&1�R'1Ì�R'15e'1Ѐ} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–Êk7 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T”›ÛÉ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–Êk7 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T”›ÛÉ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–Êk7  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T”›ÛÉ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1­©NnÚj—K bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1  `V–ê. bÒ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1<  VÖÙY*ª’þ~KËÒR¦&zúNÎê*"¢ö6ŠÊ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>/?/I¦I¦ST^‘^‘his]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIWJWLÎLÎNEOEQ¼Q¼S3T3VªVªX!Y![˜[˜]^`†`†býbýdtetgëgëibjblÙlÙnMoMq¿q¿s1t1v£v£xy{‡{‡}ù}ùk€k‚݂݄O…O‡©‡©‰ŠŒ�Œ�ŽÿŽÿ�q‘q“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxi!1S}!1èS}!1ñ½!1dñ½!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!1ç!1@ç!1½'"1¼½'"1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1ñP"1˜ñP"1�‘"1�‘"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"1‹Ú"1Ì ‹Ú"1ëþ"1H ëþ"1-,#1h-,#1µ:#1Dµ:#1qo#1Ðqo#1�#1ô�#1›#1 ›#1™$1¬™$1qò$1pqò$1«Þ%1t«Þ%1%ì%1T%ì%1W€&1àW€&1'R'1$'R'1Ïd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations15i´€ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Cƒ5uÇ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations15i´€ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Cƒ5uÇ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations15i´€  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Cƒ5uÇ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¦Æóv– bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1jª�ÍT¼üd¤  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1djêTÔ>¾(¨òr©)’`àü|—æfÐP¸8  ˆ  p ð X Ø @ À ý = ( ¨ �ú:¤äLŒô4 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚNNÂÂ)6*64‘4‘>?IŽIŽST^z^zhìhìr^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""šš‚‚ööj j"Þ"Þ$R%R'Æ'Æ):*:,®,®."/"1–1–3 4 6~6~8ò8ò:f;f=Á=Á?3@3B¥B¥DEG‰G‰IJL†L†NøNøPjQjSÜSÜUNVNXÀXÀZ2[2]¤]¤_`b�b�degtgtiæiækXlXnÊnÊp‡>‰¯‰¯‹ Œ ޑޑ�‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxi!1~!1è~!1¥¾!1d¥¾!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!1Ùç!1@Ùç!1w("1¼w("1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1«Q"1˜«Q"1I’"1I’"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ûÓ"1ð ûÓ"1=Ú"1È =Ú"1�þ"1D �þ"1Û,#1dÛ,#14#1<4#1«w#1¸«w#1…#1Ü…#1}�#1¸}�#1�$1D�$1Yò$1Yò$1“Þ%1 “Þ%1 ì%1ì ì%1c€&1xc€&13R'1¼3R'1Ûd'1À€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—Ëm9 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•›ÛÊ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—Ëm9 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•›ÛÊ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—Ëm9  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•›ÛÊ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �‰#C¶þ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 5uªî™Ý bÒ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1<5µêjŸTÔ ‰„¾>¿?sóxø(¨1±ÝR’c£ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ö ÖKKÀÀ)5*54ª4ª>!?!I–I–S T ^�^�hóhórLsL} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÖÖKKÀÀ 5 5 ª ª””  ~~óóhhÝÝR R"Ç"Ç$<%<'±'±)&*&,›,›./1…1…3ú3ú5o6o8ä8ä:Y;Y=Î=Î?C@CB¸B¸D-E-G¢G¢IJLŽLŽNOQxQxSíSíUbVbX×X×ZL[L]Á]Á_6`6b«b«d e g•g•i j llnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†P‡P‰Â‰Â‹4Œ4ަަ�‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1Y}!1èY}!1÷½!1d÷½!1eÅ!1leÅ!1ïÎ!1èïÎ!1%Ò!1R%Ò!1%ç!1‚%ç!1Ã'"1’Ã'"17/"1º7/"1Á8"1ÊÁ8"1÷;"1z÷;"1÷P"1ª÷P"1•‘"1º •‘"1 ™"1â  ™"1“¢"1ò “¢"1É¥"1¢É¥"1 Ó"1Ò Ó"1aÚ"1¢aÚ"1Áþ"1²Áþ"1,#1 ,#1I4#1ÚI4#1ów#1êów#1¹Š#1J"¹Š#1·•#1$·•#1Ÿ$1J%Ÿ$1wò$1ê&wò$1±Þ%1 )±Þ%1+ì%1â*+ì%1ù€&1,ù€&1ÉR'1²0ÉR'1qe'1Ò2€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations16j·ƒ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T5uÓ bÊwBoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*Q*2reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements11521152zinput1Elementsbinvocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations16j·ƒ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T5uÓ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations16j·ƒ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T5uÓ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 }�Aa*J bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1  MX˜ð4 bÒ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1< �XØÆF0°ÿˆ8¸à`¸8�hè@ÀX°ðž Þ  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Û ÛTTÍÍ)F*F4¿4¿>8?8I¯I¯S$T$^—^—h i sese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÛÛTTÍÍ F F ¿ ¿88±±**££•• ‡ ‡"#%y%y'ò'ò)k*k,ä,ä.]/]1Ö1Ö3O4O6È6È8A9A;º;º=3>3@¬@¬B%C%EžEžGHJ�J�LMO~O~QõQõSlTlVãVãXZYZ[Ñ[Ñ]H^H`¿`¿b4c4e©e©ghj“j“lmo{o{qîqîsatavÔvÔxGyG{º{º}-~-€ € ‚ƒ…†…†‡á‡á‰SŠSŒÅŒÅŽ7�7‘©‘©“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l%i!1~!1Ð~!1Ÿ¾!1LŸ¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1Óç!1(Óç!1q("1¤q("1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1¥Q"1€¥Q"1C’"1üC’"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1sÚ"1´ sÚ"1Óþ"18 Óþ"1,#1X,#1c2#14c2#1é„#1 é„#1·$1,·$1�ò$1ð�ò$1ÉÞ%1ôÉÞ%1í€&1Ôí€&1½R'1½R'1ee'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‹¿`, b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _ŸØÜ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‹¿`, b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _ŸØÜ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‹¿`,  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _ŸØÜ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1¢žÈXUuz. bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1S€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1“ÊJ�8¸Ë  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á á[[ÑÑ)G*G4½4½>3?3I§I§ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±++¥¥ ™ ™"#%‰%‰'ÿ'ÿ)u*u,ë,ë.a/a1×1×3M4M6Ã6Ã8999;¯;¯=%>%@›@›BCE‡E‡GýGýIsJsLçLçN[O[QÏQÏSCTCV·V·X+Y+[Ÿ[Ÿ]^`‡`‡bûbûdoeogãgãiWjWlËlËn?o?q³q³s't'v›v›xy{ƒ{ƒ}÷}÷k€k‚߂߄S…S‡LJlj;Š;Œ¯Œ¯Ž#�#‘—‘—“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1S}!1èS}!1ñ½!1dñ½!1eÅ!1leÅ!1ïÎ!1èïÎ!1%Ò!1´%Ò!1%ç!1@%ç!1Ã'"1¼Ã'"17/"1Ä7/"1Á8"1@Á8"1÷;"1 ÷;"1÷P"1˜÷P"1•‘"1•‘"1 ™"1  ™"1“¢"1˜ “¢"1É¥"1d É¥"1 Ó"1ð  Ó"1yÚ"1Ì yÚ"1Ùþ"1H Ùþ"1,#1h,#1Ù4#1DÙ4#1ƒx#1Àƒx#17‹#1ä7‹#15–#1À5–#1‡$1L‡$1_ò$1_ò$1™Þ%1™Þ%1ì%1ôì%1‡€&1€‡€&1WR'1ÄWR'1ÿd'1È€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations16jµ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U5uÒ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations16jµ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U5uÒ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations16jµ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U5uÒ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �½6`€ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 5uªî†Æ bÒ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1<5µêjŸTÔ ‰ñq¾>ª*sócã(¨œÝR’ W  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÇÇ)>*>4µ4µ>,?,I¡I¡ST^…^…höhörOsO} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbäbädWeWgÊgÊi=j=l°l°n!o!q’q’stvtvtxåxåzV{V}Ç}Ç8€8‚©‚©„…‡s‡s‰ä‰ä‹UŒUŽÆŽÆ�7‘7“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l%i!1~!1Ð~!1Ÿ¾!1LŸ¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Óç!1(Óç!1q("1¤q("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1¥Q"1€¥Q"1C’"1üC’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ô"1Ø Ô"1aÚ"1° aÚ"1Áþ"1, Áþ"1ÿ,#1Lÿ,#1Ý3#1$Ý3#1a…#1�a…#1Ÿ$1Ÿ$1wò$1àwò$1±Þ%1ä±Þ%1u€&1Äu€&1ER'1ER'1íd'1 €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀb. b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ` ÚÝ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀb. b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ` ÚÝ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀb.  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ` ÚÝ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‚~ è@` bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ŠÎ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ïo¦&]Ý”B‚ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝUUÉÉ)=*=4±4±>%?%I—I—S T ^{^{híhír_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%���� } }"õ"õ$i%i'Ý'Ý)Q*Q,Å,Å.9/91­1­3!4!6•6•8 9 ;};}=ñ=ñ?e@eBÙBÙDMEMGÁGÁI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnãnãpUqUsÇsÇu9v9x«x«z{}�}�€‚s‚s„å„å†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1_}!1Ð_}!1ý½!1Lý½!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1+ç!1(+ç!1É'"1¤É'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1ýP"1€ýP"1›‘"1ü›‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1W2#1,W2#1׊#1˜׊#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1)�&1Ì)�&1ùR'1ùR'1¡e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations17k¸„ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 9y7wà  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations17k¸„ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 9y7wà  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations17k¸„  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 9y7wà   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 t”8Tt bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1„¿?zú5µ´ ô  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚRRÊÊ)B*B4º4º>2?2I¨I¨ST^”^”h i s€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""ššŠŠ z z"ò"ò$j%j'â'â)Z*Z,Ò,Ò.J/J1Â1Â3:4:6²6²8*9*;¢;¢=>@’@’B C E‚E‚GúGúIrJrLèLèN^O^QÔQÔSJTJVÀVÀX6Y6[¬[¬]"^"`˜`˜bce„e„gúgúipjplælæn\o\qÒqÒsHtHv¾v¾x4y4{ª{ª} ~ €–€–‚ ƒ …‚…‚‡ø‡ø‰nŠnŒäŒäŽZ�Z‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1û}!1Ðû}!1™¾!1L™¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Çç!1(Çç!1e("1¤e("1[/"1¬[/"1å8"1(å8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1 Ú"1´  Ú"1mþ"18 mþ"1¯+#1X¯+#1]2#14]2#1Ñ„#1 Ñ„#1¥$1,¥$1}ò$1ð}ò$1·Þ%1ô·Þ%1u€&1Ôu€&1ER'1ER'1íd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´U! b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 jª›Ûî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´U! b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 jª›Ûî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´U!  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 jª›Ûî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥Å—“¶F] bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1F€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1†»;pð%¥®î €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94®4®>!?!I”I”ST^z^zhíhír`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?M@MBÀBÀD3E3G¦G¦IJLŒLŒNÿNÿPrQrSåSåUXVXXËXËZ>[>]±]±_$`$b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†T‡T‰ƉƋ8Œ8ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1Y}!1ÐY}!1÷½!1L÷½!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1+ç!1(+ç!1É'"1¤É'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1ýP"1€ýP"1›‘"1ü›‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1Ó4#1,Ó4#1O‹#1˜O‹#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1·€&1Ì·€&1‡R'1‡R'1/e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations17k¶‚ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 8x6vÞ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations17k¶‚ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 8x6vÞ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations17k¶‚  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 8x6vÞ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ”´;[Jj bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¿€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ðp«+fæ!¡/ o  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÄÄ):*:4°4°>&?&IšIšST^‚^‚höhörjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$L%L'Â'Â)8*8,®,®.$/$1š1š346†6†8ü8ü:r;r=è=è?^@^BÔBÔDJEJGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqvqvsêsêu^v^xÒxÒzF{F}º}º.€.‚¢‚¢„…‡ЇЉþ‰þ‹rŒr޿޿�Z‘Z“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1û}!1Ðû}!1™¾!1L™¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Íç!1(Íç!1k("1¤k("1[/"1¬[/"1å8"1(å8"1<"1ô<"1ŸQ"1€ŸQ"1=’"1ü=’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1+Ô"1Ø +Ô"1=Ú"1° =Ú"1�þ"1, �þ"1Û,#1LÛ,#1×3#1$×3#1C…#1�C…#1�$1�$1eò$1àeò$1ŸÞ%1äŸÞ%1€&1Ä€&1åQ'1åQ'1�d'1 €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µW# b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 k«›Ûð0 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µW# b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 k«›Ûð0 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µW#  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 k«›Ûð0  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 wsŠÒ: bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1{¿€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÚZ�DÄùy#c €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚMMÀÀ)3*34¦4¦>?IˆIˆSùSù]j^jhÛhÛrLsL} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠýýppããV V"É"É$<%<'¯'¯)"*",•,•./1{1{3î3î5a6a8Ô8Ô:G;G=º=º?+@+BœBœD E G~G~IïIïK`L`NÑNÑPBQBS³S³U$V$X•X•Z[]w]w_è_èaYbYdÊdÊf;g;i¬i¬klnŽnŽpÿpÿrpspuáuáwRxRzÃzÃ|4}4¥¥�‚„‡„‡†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1_}!1Ð_}!1ý½!1Lý½!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!11ç!1(1ç!1Ï'"1¤Ï'"1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1Q"1€Q"1¡‘"1ü¡‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1W2#1,W2#1U‹#1˜U‹#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1ó€&1Ìó€&1ÃR'1ÃR'1ke'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l¹… b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IM�@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l¹… b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IM�@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l¹…  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IM�@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 k‹&Oo bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÂB„FÆ0 p  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÇÇ)>*>4µ4µ>,?,I¡I¡ST^‹^‹hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbæbæd[e[gÐgÐiEjElºlºn/o/q¤q¤stvŽvŽxy{x{x}í}íb€b‚ׂׄL…L‡Á‡Á‰6Š6Œ«Œ«Ž � ‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1õ}!1Ðõ}!1“¾!1L“¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Áç!1(Áç!1_("1¤_("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1õÙ"1´ õÙ"1Uþ"18 Uþ"1—+#1X—+#1]2#14]2#1Ñ„#1 Ñ„#1·$1,·$1�ò$1ð�ò$1ÉÞ%1ôÉÞ%1À&1ÔÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©J b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 jªÐâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©J b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 jªÐâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©J  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 jªÐâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥ÅŒˆ¤4@ô bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1„¹9nî#£¬ì €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßTTÉÉ)>*>4³4³>)?)I�I�ST^…^…hùhùroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ‰‰þþs s"è"è$]%]'Ò'Ò)G*G,¼,¼.1/11¦1¦346�6�89;z;z=ï=ï?d@dBÙBÙDNENGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒyŒyŽîŽî�c‘c“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1_}!1Ð_}!1ý½!1Lý½!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!11ç!1(1ç!1Ï'"1¤Ï'"1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1Q"1€Q"1¡‘"1ü¡‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1Ù4#1,Ù4#1Ù‹#1˜Ù‹#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l·ƒ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JLŒþ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l·ƒ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JLŒþ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l·ƒ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JLŒþ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‹«>^4T bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1‚€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ˆÊJŒ Nβ ò  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙOOÄÄ)9*94®4®>#?#I–I–S T ^|^|hïhïrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰ÿÿuuëëa a"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnänäpWqWsÊsÊu=v=x°x°z#{#}–}– € ‚|‚|„ï„ï†b‡b‰Õ‰Õ‹HŒH޻޻�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1õ}!1Ðõ}!1“¾!1L“¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1Çç!1(Çç!1e("1¤e("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1%Ô"1Ø %Ô"1gÚ"1° gÚ"1Çþ"1, Çþ"1-#1L-#1×3#1$×3#1I…#1�I…#1«$1«$1ƒò$1àƒò$1½Þ%1ä½Þ%1{€&1Ä{€&1KR'1RKR'1ód'1r€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªL b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 k«Ñä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªL b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 k«Ñä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªL  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 k«Ñä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 lht¼?_ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1y½€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ØX� BÂ÷w!a €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛPPÅÅ):*:4¯4¯>$?$I—I—S T ^{^{híhírasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ‚operation.6exchangezOperatorKernel107 :z2breasonOperatorInputsvalue8invocations1ÇðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹uuêê_ _"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I$J$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b¡b¡deg…g…i÷i÷kilinÛnÛpMqMs¿s¿u1v1x£x£z{}‡}‡ùù�k‚k„݄݆Q‡Q‰ĉċ7Œ7ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1e}!1Ðe}!1¾!1L¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!17ç!1(7ç!1Õ'"1¤Õ'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1 Q"1€ Q"1§‘"1ü§‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1Ú"1´ Ú"1yþ"10 yþ"1»+#1P»+#1]2#1,]2#1 Œ#1˜ Œ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Õ€&1ÌÕ€&1¥R'1¥R'1Me'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19mº† b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 .n5uÉ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19mº† b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 .n5uÉ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19mº†  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 .n5uÉ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 b‚¥Åõ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÍMšgçà  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff à àVVÌÌ)B*B4¸4¸>,?,I I ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffààZZÔÔ N N È ÈBB¼¼66°°&&œœ ˆ ˆ"þ"þ$t%t'ê'ê)`*`,Ö,Ö.L/L1Â1Â38486®6®8$9$;š;š=>@„@„BøBøDlElGàGàITJTLÈLÈN3?3IªIªS!T!^“^“hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâI[J[LÒLÒNIOIQÀQÀS7T7V®V®X%Y%[œ[œ]^`Š`ŠbcexexgïgïifjflÝlÝnOoOqÁqÁs3t3v¥v¥xy{‰{‰}û}ûm€m‚߂߄Q…Q‡ÇÉ5Š5Œ§Œ§Ž�‘‹‘‹“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1e}!1Ðe}!1¾!1L¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!17ç!1(7ç!1Õ'"1¤Õ'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1 Q"1€ Q"1§‘"1ü§‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1CÚ"1´ CÚ"1£þ"10 £þ"1å+#1På+#1ë4#1,ë4#1™Œ#1˜™Œ#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1c€&1Ìc€&13R'13R'1Ûd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¸„ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 -m5uÊ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¸„ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 -m5uÊ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¸„  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 -m5uÊ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‚¢¥Å> bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Í€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations14´�ÎN›m ­  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞRRÆÆ):*:4®4®> ? I’I’ST^v^vhèhèrZsZ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žž††úún n"â"â$V%V'Ê'Ê)>*>,²,².&/&1š1š346‚6‚8ö8ö:j;j=Þ=Þ?P@PBÂBÂD4E4G¦G¦IJLŠLŠNüNüPnQnSàSàURVRXÄXÄZ6[6]¨]¨_`bŒbŒdþdþfpgpiâiâkTlTnÆnÆp8q8sªsªuvxŽxŽz{}r}rää�V‚V„ȄȆ:‡:‰«‰«‹ŒŽ�Ž��þ�þ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ï}!1Ðï}!1�¾!1L�¾!1eÅ!1TeÅ!1ïÎ!1ÐïÎ!1%Ò!1œ%Ò!1»ç!1(»ç!1Y("1¤Y("17/"1¬7/"1Á8"1(Á8"1÷;"1ô÷;"1�Q"1€�Q"1+’"1ü+’"1 ™"1  ™"1“¢"1€ “¢"1É¥"1L É¥"1Ô"1Ø Ô"1sÚ"1° sÚ"1Óþ"1, Óþ"1-#1L-#1Ý3#1$Ý3#1……#1�……#1±$1±$1‰ò$1à‰ò$1ÃÞ%1äÃÞ%1“€&1Ä“€&1cR'1cR'1 e'1 €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R’Ç à bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R’Ç à bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R’Ç à  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 a]^¦Ff bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1|À€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ðp¬,hè$¤\œ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ú ÚOOÄÄ)9*94®4®>%?%IšIšST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÚÚOOÄÄ 9 9 ® ®##˜˜  ‚‚÷÷llááV V"Ë"Ë$@%@'µ'µ)***,Ÿ,Ÿ./1‰1‰3þ3þ5s6s8è8è:];]=Ò=Ò?G@GB¼B¼D1E1G¦G¦IJL’L’NOQ|Q|SñSñUfVfXÛXÛZP[P]Å]Å_:`:b¯b¯d$e$g™g™ijlƒlƒnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1k}!1Ðk}!1 ¾!1L ¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1=ç!1(=ç!1Û'"1¤Û'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1Q"1€Q"1­‘"1ü­‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1 Ú"1´  Ú"1mþ"10 mþ"1¯+#1P¯+#1c2#1,c2#1½Œ#1˜½Œ#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1)�&1Ì)�&1ùR'1ùR'1¡e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n»‡ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @6vÍ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n»‡ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @6vÍ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n»‡  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @6vÍ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¥ÅYyâ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ØX°0ˆ� Ð  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff ß ßXXÑÑ)J*J4Ã4Ã>7@°@°B)C)E¢E¢GHJ”J”L M O‚O‚QùQùSpTpVçVçX^Y^[Õ[Õ]L^L`Ã`Ãb8c8e­e­g"h"j—j—l m ooqòqòsetevØvØxKyK{¾{¾}1~1€¤€¤‚ƒ…ЅЇý‡ý‰oŠoŒáŒáŽS�S‘őœ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1ã}!1Ðã}!1�¾!1L�¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1µç!1(µç!1S("1¤S("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1‡Q"1€‡Q"1%’"1ü%’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1ËÙ"1´ ËÙ"1+þ"18 +þ"1m+#1Xm+#1c2#14c2#1g…#1 g…#1É$1,É$1¡ò$1ð¡ò$1ÛÞ%1ôÛÞ%1ÿ€&1Ôÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“4 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V–Ë Ç bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“4 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V–Ë Ç bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“4  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V–Ë Ç  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥Åvr€Ä bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÇGŽUÕ8x €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk å å__ÕÕ)K*K4Á4Á>7?7I«I«ST^“^“his{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkåå__ÙÙ S S Í ÍGGÁÁ;;µµ//©©## � �"#%�%�'(*y*y,ï,ï.e/e1Û1Û3Q4Q6Ç6Ç8=9=;³;³=)>)@Ÿ@ŸBCE‹E‹GHJwJwLëLëN_O_QÓQÓSGTGV»V»X/Y/[£[£]^`‹`‹bÿbÿdsesgçgçi[j[lÏlÏnCoCq·q·s+t+vŸvŸxy{‡{‡}û}ûo€o‚ã‚ã„W…W‡ˇˉ?Š?Œ³Œ³Ž'�'‘›‘›“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1k}!1Ðk}!1 ¾!1L ¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!17ç!1(7ç!1Õ'"1¤Õ'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1 Q"1€ Q"1§‘"1ü§‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"11Ú"1´ 1Ú"1‘þ"10 ‘þ"1Ó+#1PÓ+#1÷4#1,÷4#1S�#1˜S�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n¹… b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @6vÎ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n¹… b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @6vÎ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n¹…  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @6vÎ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¥Åy™( bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1˜Ø€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`à8¸�èh(h €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ý ÝTTËË)B*B4¹4¹>0?0I¥I¥ST^‰^‰húhúrlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÝÝTTËË B B ¹ ¹00§§••  ƒƒúúq q"è"è$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n%o%q–q–stvxvxxéxézZ{Z}Ë}Ë<€<‚­‚­„…‡�‡�‰ŠŒrŒrŽãŽã�T‘T“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1é}!1Ðé}!1‡¾!1L‡¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1µç!1(µç!1S("1¤S("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1‡Q"1€‡Q"1%’"1ü%’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ô"1Ø Ô"1sÚ"1° sÚ"1Óþ"1, Óþ"1-#1L-#1ï3#1$ï3#1ë…#1�ë…#1±$1±$1‰ò$1à‰ò$1ÃÞ%1äÃÞ%1‡€&1ć€&1WR'1WR'1ÿd'1 €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”6 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•Ê Ç bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”6 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•Ê Ç bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”6  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•Ê Ç  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 VRRšGg bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1‡Ë€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1œãcª*qñ¿ÿ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áYYÍÍ)A*A4µ4µ>)?)I›I›S T ^^hñhñrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‘‘  � �"ù"ù$m%m'á'á)U*U,É,É.=/=1±1±3%4%6™6™8 9 ;�;�=õ=õ?i@iBÝBÝDQEQGÅGÅI9J9L«L«NOQ�Q�STVsVsXåXåZW[W]É]É_;`;b­b­deg‘g‘ijlulunçnçpYqYsËsËu=v=x¯x¯z!{!}“}“€‚w‚w„é„é†[‡[‰͉͋?Œ?ޱޱ�#‘#“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1q}!1Ðq}!1¾!1L¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Cç!1(Cç!1á'"1¤á'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1Q"1€Q"1³‘"1ü³‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1ûÙ"1´ ûÙ"1[þ"10 [þ"1�+#1P�+#1c2#1,c2#1q�#1˜q�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1A�&1ÌA�&1S'1S'1¹e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¼ˆ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 #c5uÏ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¼ˆ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 #c5uÏ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¼ˆ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 #c5uÏ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¥ÅPpÐð bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ãcÆF©)@€ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÎÎ)F*F4¾4¾>6?6I¬I¬S"T"^˜^˜his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$n%n'æ'æ)^*^,Ö,Ö.N/N1Æ1Æ3>4>6¶6¶8.9.;¦;¦=>@–@–BCE†E†GþGþIvJvLìLìNbObQØQØSNTNVÄVÄX:Y:[°[°]&^&`œ`œbceˆeˆgþgþitjtlêlên`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1Ý}!1ÐÝ}!1{¾!1L{¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1¯ç!1(¯ç!1M("1¤M("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1�Q"1€�Q"1’"1ü’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1yÚ"1´ yÚ"1Ùþ"18 Ùþ"1,#1X,#1c2#14c2#1Á…#1 Á…#1·$1,·$1�ò$1ð�ò$1ÉÞ%1ôÉÞ%1�€&1Ô�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆ)õ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™Îªê bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆ)õ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™Îªê bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆ)õ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™Îªê  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1kgnþHhþ² bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÒR¤$vö�Ð €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞSSÈÈ)=*=4²4²>%?%I˜I˜S T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽxxííb b"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?Q@QBÄBÄD7E7GªGªIJL�L�NOQvQvSéSéU\V\XÏXÏZB[B]µ]µ_(`(b›b›deg�g�iôiôkglgnÚnÚpMqMsÀsÀu3v3x¦x¦z{}Œ}Œÿÿ�r‚r„å„å†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1q}!1Ðq}!1¾!1L¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1=ç!1(=ç!1Û'"1¤Û'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1Q"1€Q"1­‘"1ü­‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1Ú"1´ Ú"1þ"10 þ"1Á+#1PÁ+#1 5#1, 5#1Ž#1˜Ž#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;oº† b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 "b5uÐ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;oº† b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 "b5uÐ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;oº†  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 "b5uÐ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¥År’ô bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1£ã€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Œ oïRÒ5µã# €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜRRÈÈ)>*>4´4´>*?*IžIžST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒxxîîd d"Ú"Ú$P%P'Æ'Æ)<*<,²,².(/(1ž1ž346Š6Š89;v;v=ì=ì?b@bBØBØDNENGÄGÄI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1Ý}!1ÐÝ}!1{¾!1L{¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1¯ç!1(¯ç!1M("1¤M("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1�Q"1€�Q"1’"1ü’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ô"1Ø Ô"1Ú"1° Ú"1ßþ"1, ßþ"1-#1L-#1û3#1$û3#1Q†#1�Q†#1¥$1¥$1}ò$1à}ò$1·Þ%1ä·Þ%1'€&1Ä'€&1÷Q'1÷Q'1Ÿd'1 €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰+÷ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 X˜Í «ë bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰+÷ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 X˜Í «ë bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰+÷  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 X˜Í «ë  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 KGG�Kk bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1’Ö€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1HÈšìl¾>"b €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞQQÄÄ)7*74ª4ª>?IŒIŒSýSý]n^nhßhßrPsP} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽttççZ Z"Í"Í$@%@'³'³)&*&,™,™. / 113ò3ò5e6e8Ø8Ø:K;K=¾=¾?/@/B B DEG‚G‚IóIóKdLdNÕNÕPFQFS·S·U(V(X™X™Z [ ]{]{_ì_ìa]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ü†üˆl‰l‹Ü‹Ü�LŽL�¼�¼’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1é}!1�é}!1‡¾!1‡¾!1­Â!1$­Â!1»ç!1�»ç!1Y("1Y("1,"1$,"1�Q"1��Q"1+’"1+’"1Q–"1$Q–"1áØ"1�áØ"1Aý"1üAý"1)0#1 )0#1õ~#1| õ~#1á$1Ü á$1¹ò$1  ¹ò$1óÞ%1¤ óÞ%1�&1„ �&1íR'1ÈíR'1•e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyM�Κ3ÿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyM�Κ3ÿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyM�Κ3ÿ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥Å  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww î îeeÜÜ)S*S4Ê4Ê>???I´I´S)T)^ž^žhisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwîîeeÜÜ S S Ê ÊAA¸¸//¦¦””  ‚ ‚"ù"ù$p%p'ç'ç)^*^,Õ,Õ.L/L1Ã1Ã3:4:6±6±8(9(;Ÿ;Ÿ=>@�@�BCE{E{GòGòIgJgLÜLÜNQOQQÆQÆS;T;V°V°X%Y%[š[š]^`„`„bùbùdnengãgãiXjXlÍlÍnBoBq·q·s,t,v¡v¡xy{‹{‹}~€u€u‚ê‚ê„_…_‡Ô‡Ô‰IŠIŒ¾Œ¾Ž3�3‘¨‘¨“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1×}!1�×}!1u¾!1 u¾!1}Å!1}Å!1©ç!1€©ç!1G("1üG("1O/"1O/"1{Q"1p{Q"1’"1ì’"1!™"1ô!™"1Ú"1`Ú"1ßþ"1Ìßþ"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1á€&1T á€&1±R'1˜±R'1Ye'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|é b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ó€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|é b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ó€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|é b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ó €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Dd  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ì ìaaÖÖ)K*K4À4À>4?4I¨I¨ST^�^�hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttììddÜÜ T T Ì ÌDD¼¼44¬¬!!––  € €"õ"õ$j%j'ß'ß)T*T,É,É.>/>1³1³3(4(6�6�89;‡;‡=ü=ü?q@qBæBæD[E[GÐGÐIDJDL¸L¸N,O,Q Q STVˆVˆXüXüZp[p]ä]ä_X`XbÌbÌd@e@g´g´i(j(lœlœnoq„q„søsøulvlxàxàzT{T}È}È<€<‚°‚°„$…$‡˜‡˜‰ Š Œ‚Œ‚Ž÷Ž÷�l‘l“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1ï}!1�ï}!1�¾!1�¾!1­Â!1$­Â!1»ç!1�»ç!1Y("1Y("1,"1$,"1�Q"1��Q"1+’"1+’"1Q–"1$Q–"1áØ"1�áØ"1Aý"1üAý"1)0#1 )0#1õ~#1| õ~#1Ï$1Ü Ï$1§ò$1  §ò$1áÞ%1¤ áÞ%1±€&1„ ±€&1�R'1È�R'1)e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzKÊ–6 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzKÊ–6 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzKÊ–6 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥Å  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ì ìbb××)L*L4Á4Á>4?4I§I§ST^�^�hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvììbbØØ N N Ä Ä::°°&&œœˆˆþþt t"ê"ê$_%_'Ô'Ô)I*I,¾,¾.3/31¨1¨346’6’89;|;|=ñ=ñ?f@fBÛBÛDPEPGÅGÅI8J8L«L«NOQ‘Q‘STVwVwXêXêZ][]]Ð]Ð_C`Cb¶b¶d)e)gœgœijl‚l‚nõnõphqhsÛsÛuNvNxÁxÁz4{4}§}§€‚�‚�„…‡s‡s‰æ‰æ‹YŒYŽÌŽÌ�?‘?“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1×}!1�×}!1u¾!1 u¾!1}Å!1}Å!1©ç!1€©ç!1G("1üG("1O/"1O/"1{Q"1p{Q"1’"1ì’"1!™"1ô!™"1…Ú"1`…Ú"1åþ"1Ìåþ"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{é b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ó€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{é b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ó€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{é b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ó €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Mm  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ê ê__ÔÔ)I*I4¾4¾>1?1I¤I¤ST^ˆ^ˆhúhúrlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttêê``ÖÖ L L  Â88®®$$šš„„ùùn n"ã"ã$X%X'Í'Í)B*B,·,·.,/,1¡1¡346‹6‹89;u;u=ê=ê?_@_BÔBÔDIEIG¾G¾I1J1L¤L¤NOQŠQŠSýSýUpVpXãXãZV[V]É]É_<`v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰ωϋBŒB޵޵�(‘(“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1ƒ}!1�ƒ}!1!¾!1 !¾!1ËÂ!1ËÂ!1Uç!1€Uç!1ó'"1üó'"1�,"1�,"1'Q"1p'Q"1Å‘"1ìÅ‘"1o–"1ôo–"1yÚ"1`yÚ"1Ùþ"1ÌÙþ"1#0#1ì#0#1ï~#1L ï~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1í€&1T í€&1½R'1˜½R'1ee'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ë— b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1#c€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ë— b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1#c€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ë— b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1#c €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ee  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww ñ ñggÝÝ)S*S4Ç4Ç>;?;I¯I¯S#T#^—^—h i s}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwññkkåå _ _ Ù ÙSSÍÍGGÁÁ77­­## ™ ™"#%…%…'û'û)q*q,ç,ç.]/]1Ó1Ó3I4I6¿6¿8595;«;«=>@“@“BCE{E{GïGïIcJcL×L×NKOKQ¿Q¿S3T3V§V§XY[�[�]^`w`wbëbëd_e_gÓgÓiGjGl»l»n/o/q£q£stv‹v‹xÿxÿzs{s}ç}ç[€[‚ςτC…C‡µ‡µ‰(Š(Œ›Œ›Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Ý}!1�Ý}!1{¾!1 {¾!1SÅ!1SÅ!1¯ç!1€¯ç!1M("1üM("1%/"1%/"1�Q"1p�Q"1’"1ì’"1÷˜"1ô÷˜"1sÚ"1`sÚ"1Óþ"1ÌÓþ"1#0#1ì#0#1ï~#1L ï~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…&ò b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ªê€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…&ò b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ªê€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…&ò b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ªê €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1=]  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp ç ç^^ÕÕ)L*L4Ã4Ã>:?:I±I±S(T(^š^šh i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppçç^^ÕÕ L L à Ã::±±((ŸŸ�� { {"ò"ò$i%i'à'à)W*W,Î,Î.E/E1¼1¼33436ª6ª8!9!;˜;˜=>@†@†BýBýDtEtGëGëIbJbLÙLÙNPOPQÇQÇS>T>VµVµX,Y,[£[£]^`‘`‘bceegögöimjmlälänVoVqÈqÈs:t:v¬v¬xy{�{�}~€t€t‚æ‚æ„X…X‡ʇʉ<Š<Œ®Œ®Ž � ‘’‘’“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1}}!1�}}!1¾!1 ¾!1ËÂ!1ËÂ!1Oç!1€Oç!1í'"1üí'"1�,"1�,"1!Q"1p!Q"1¿‘"1ì¿‘"1o–"1ôo–"1…Ú"1`…Ú"1åþ"1Ìåþ"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1u€&1T u€&1ER'1˜ER'1íd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ç“ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1"b€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ç“ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1"b€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ç“ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1"b €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Nn  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu í íaaÕÕ)I*I4»4»>-?-IŸIŸST^ƒ^ƒhõhõrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuííeeÝÝ U U Í ÍEE½½55­­!!••  } }"ñ"ñ$e%e'Ù'Ù)M*M,Á,Á.5/51©1©346‘6‘89;y;y=ë=ë?]@]BÏBÏDAEAG³G³I%J%L—L—N O Q{Q{SíSíU_V_XÑXÑZC[C]µ]µ_'`'b™b™d e g}g}iïiïkalanÓnÓpEqEs·s·u)v)x›x›z { }}ññ�c‚c„Õ„Õ†E‡E‰¶‰¶‹'Œ'Ž˜Ž˜� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Ý}!1�Ý}!1{¾!1 {¾!1SÅ!1SÅ!1¯ç!1€¯ç!1M("1üM("1%/"1%/"1�Q"1p�Q"1’"1ì’"1÷˜"1ô÷˜"1…Ú"1`…Ú"1åþ"1Ìåþ"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„&ò b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1«ë€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„&ò b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1«ë€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„&ò b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1«ë €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Nn  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp å åZZÏÏ)D*D4¹4¹>.?.I£I£ST^Š^Šhühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppååZZÏÏ D D ¹ ¹..££��wwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD^  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss ì ìeeÞÞ)W*W4Ð4Ð>G?GI¾I¾S3T3^¦^¦hisŠsŠ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssììeeÞÞ W W Ð ÐIIÂÂ;;´´--¦¦ ˜ ˜"#%Š%Š'(*|*|,õ,õ.n/n1ç1ç3`4`6Ù6Ù8R9R;Ë;Ë=D>D@½@½B6C6E¯E¯G(H(JŸJŸLMO�O�QRT{T{VòVòXiYi[à[à]W^W`Î`ÎbCcCe¸e¸g-h-j¢j¢lmoŠoŠqýqýsptpvãvãxVyV{É{É}<~<€¯€¯‚"ƒ"…•…•‡ˆŠxŠxŒêŒêŽ\�\‘ΑΓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ã}!1�ã}!1�¾!1 �¾!1}Å!1}Å!1µç!1€µç!1S("1üS("1O/"1O/"1‡Q"1p‡Q"1%’"1ì%’"1!™"1ô!™"1mÚ"1`mÚ"1Íþ"1ÌÍþ"10#1ì0#1é~#1L é~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1 �&1T  �&1ÛR'1˜ÛR'1ƒe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ/û b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ/û b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ/û b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations18X  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp ê êddÚÚ)P*P4Æ4Æ>:?:I®I®S"T"^–^–h i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppêêddÞÞ X X Ò ÒLLÆÆ@@ºº44®®(( ¢ ¢"#%’%’'(*~*~,ô,ô.j/j1à1à3V4V6Ì6Ì8B9B;¸;¸=.>.@¤@¤BCE�E�GHJzJzLîLîNbObQÖQÖSJTJV¾V¾X2Y2[¦[¦]^`Ž`Žbcevevgêgêi^j^lÒlÒnFoFqºqºs.t.v¢v¢xy{Š{Š}þ}þr€r‚æ‚æ„Z…Z‡·ΉBŠBŒ¶Œ¶Ž*�*‘ž‘ž“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1w}!1�w}!1¾!1 ¾!1­Â!1­Â!1Iç!1€Iç!1ç'"1üç'"1,"1,"1Q"1pQ"1¹‘"1칑"1Q–"1ôQ–"1Ú"1`Ú"1ßþ"1Ìßþ"10#1ì0#1é~#1L é~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1À&1T À&1“R'1˜“R'1;e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÄ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÄ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÄ� b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Oo  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq è è__ÖÖ)M*M4Ä4Ä>9?9I®I®S!T!^’^’hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqèè__ÖÖ M M Ä Ä;;²²))  ŽŽ | |"ó"ó$j%j'á'á)X*X,Ï,Ï.F/F1½1½34446«6«8"9";™;™=>@‡@‡BþBþDuEuGìGìIaJaLÖLÖNKOKQÀQÀS5T5VªVªXY[”[”] ^ `~`~bñbñddedg×g×iJjJl½l½n.o.qŸqŸstv�v�xòxòzc{c}Ô}ÔE€E‚¶‚¶„'…'‡—‡—‰ŠŒyŒyŽêŽê�[‘[“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ã}!1�ã}!1�¾!1 �¾!1}Å!1}Å!1µç!1€µç!1S("1üS("1O/"1O/"1‡Q"1p‡Q"1%’"1ì%’"1!™"1ô!™"1Ú"1`Ú"1ßþ"1Ìßþ"10#1ì0#1é~#1L é~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�/û b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ú€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�/û b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ú€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�/û b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ú €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Mm  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp è è``ÔÔ)H*H4¼4¼>.?.I I ST^„^„höhörhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppèè``ØØ P P È È@@¸¸00¨¨  ˜˜ ˆ ˆ"#%t%t'è'è)\*\,Ð,Ð.D/D1¸1¸3,4,6 6 89;ˆ;ˆ=ü=ü?p@pBäBäDXEXGÌGÌI>J>L°L°N"O"Q”Q”STVxVxXêXêZ\[\]Î]Î_@`@b²b²d$e$g–g–ijlzlznìnìp^q^sÐsÐuBvBx´x´z&{&}˜}˜ € ‚|‚|„î„î†`‡`‰Ò‰Ò‹DŒD޶޶�(‘(“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1w}!1�w}!1¾!1 ¾!1×Â!1×Â!1Cç!1€Cç!1á'"1üá'"1©,"1©,"1Q"1pQ"1³‘"1쳑"1{–"1ô{–"1gÚ"1`gÚ"1Çþ"1ÌÇþ"10#1ì0#1é~#1L é~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1A�&1T A�&1S'1˜S'1¹e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÅ‘ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1:z€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÅ‘ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1:z€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÅ‘ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1:z €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations19Y  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo ç ç__××)O*O4Ç4Ç>=?=I³I³S)T)^Ÿ^Ÿhis‹s‹} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooçç__×× O O Ç Ç??··//§§—— ‡ ‡"ÿ"ÿ$w%w'ï'ï)g*g,ß,ß.W/W1Ï1Ï3G4G6¿6¿8797;¯;¯='>'@Ÿ@ŸBCE�E�GHJ}J}LóLóNiOiQßQßSUTUVËVËXAYA[·[·]-^-`£`£bce�e�ghj{j{lñlñngogqÝqÝsStSvÉvÉx?y?{µ{µ}+~+€¡€¡‚ƒ…�…�‡ˆŠyŠyŒïŒïŽe�e‘Û‘Û“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1é}!1�é}!1‡¾!1 ‡¾!1;Å!1;Å!1»ç!1€»ç!1Y("1üY("1 /"1 /"1�Q"1p�Q"1+’"1ì+’"1ߘ"1ôߘ"1ÕØ"1`ÕØ"15ý"1Ì5ý"10#1ì0#1é~#1L é~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—8 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1›Û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—8 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1›Û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—8 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1›Û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥Å  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii ß ßTTÉÉ)>*>4±4±>$?$I—I—S T ^}^}hðhðrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiißßUUËË A A · ·--££��yyîîc c"Ø"Ø$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=Ý=Ý?P@PBÃBÃD6E6G©G©IJL�L�NOQuQuSèSèU[V[XÎXÎZA[A]´]´_'`'bšbšd e g€g€ióiókflfnÙnÙpLqLs¿s¿u2v2x¥x¥z{}‹}‹þþ�q‚q„ä„ä†U‡U‰ljNj9Œ9ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1q}!1�q}!1¾!1 ¾!1ÝÂ!1ÝÂ!1Cç!1€Cç!1á'"1üá'"1¯,"1¯,"1Q"1pQ"1³‘"1쳑"1�–"1ô�–"1Ú"1`Ú"1ßþ"1Ìßþ"10#1ì0#1é~#1L é~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÁ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1;{€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÁ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1;{€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÁ� b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1;{ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Nn  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ã ãYYÏÏ)E*E4»4»>/?/I£I£ST^‹^‹hÿhÿrsss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““  õõk k"á"á$W%W'Í'Í)C*C,¹,¹.///1¥1¥346‘6‘89;};}=ó=ó?i@iBßBßDUEUGËGËI?J?L³L³N'O'Q›Q›STVƒVƒX÷X÷Zk[k]ß]ß_S`SbÇbÇd;e;g¯g¯i#j#l—l—n o qqsósóugvgxÛxÛzO{O}Ã}Ã7€7‚«‚«„…‡“‡“‰ŠŒ{Œ{ŽïŽï�c‘c“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1é}!1�é}!1‡¾!1 ‡¾!1;Å!1;Å!1»ç!1€»ç!1Y("1üY("1 /"1 /"1�Q"1p�Q"1+’"1ì+’"1ߘ"1ôߘ"1ÕØ"1`ÕØ"15ý"1Ì5ý"10#1ì0#1é~#1L é~#1Ÿ$1¬ Ÿ$1wò$1p wò$1±Þ%1t ±Þ%1!€&1T !€&1ñQ'1˜ñQ'1™d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–8 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1›Û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–8 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1›Û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–8 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1›Û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥Å  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk á áTTÇÇ):*:4«4«>?I�I�SþSþ]o^ohàhàrOsO} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkááWWÍÍ C C ¹ ¹//¥¥‘‘wwêê] ]"Ð"Ð$C%C'¶'¶))*),œ,œ./1‚1‚3õ3õ5h6h8Û8Û:N;N=¿=¿?0@0B¡B¡DEGƒGƒIôIôKeLeNÖNÖPGQGS¸S¸U)V)XšXšZ [ ]|]|_í_ía^b^dÏdÏf@g@i±i±k"l"n“n“pqsusuuæuæwWxWzÈzÈ|9}9ªª�‚„Œ„Œ†û†ûˆk‰k‹Û‹Û�KŽK�»�»’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1q}!1�q}!1¾!1 ¾!1­Â!1­Â!1=ç!1€=ç!1Û'"1üÛ'"1,"1,"1Q"1pQ"1­‘"1ì­‘"1Q–"1ôQ–"1[Ú"1`[Ú"1»þ"1Ì»þ"10#1ì0#1ã~#1L ã~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÂŽ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÂŽ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÂŽ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations12R  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk â âYYÐÐ)G*G4¾4¾>3?3I¨I¨ST^’^’his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkââYYÐÐ G G ¾ ¾55¬¬##ššˆˆÿÿv v"í"í$d%d'Û'Û)R*R,É,É.@/@1·1·3.4.6¥6¥89;“;“= > @�@�BøBøDoEoGæGæI[J[LÐLÐNEOEQºQºS/T/V¤V¤XY[Ž[Ž]^`x`xbíbídbebg×g×iLjLlÁlÁn6o6q«q«s t v•v•x y {{}ô}ôi€i‚Þ‚Þ„S…S‡ȇȉ=Š=Œ²Œ²Ž'�'‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1ï}!1�ï}!1�¾!1 �¾!1qÅ!1qÅ!1»ç!1€»ç!1Y("1üY("1C/"1C/"1�Q"1p�Q"1+’"1ì+’"1™"1ô™"1ÏØ"1`ÏØ"1/ý"1Ì/ý"10#1ì0#1ã~#1L ã~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1À&1T À&1“R'1´ “R'1;e'1"€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l A  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ø€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l A  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ø€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l A  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ø €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥Å  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àUUÊÊ)?*?4´4´>(?(IœIœST^„^„høhørlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7operation.6exchange :z2breasonOperatorInputsvalue8invocations1 b’)woperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:p122pÒêrŠš²ÂÚêreasonvalueÇðhhààXXÐÐ H H À À88°°((  ŠŠÿÿt t"é"é$^%^'Ó'Ó)H*H,½,½.2/21§1§346‘6‘89;{;{=ð=ð?e@eBÚBÚDOEOGÄGÄI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡Œ‡Œ‰ŠŒvŒvŽëŽë�`‘`“ZkernelSpecâPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })zoutputElements"640zinput0Elements*6144zinput1Elements252488binvocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1k}!1�k}!1 ¾!1  ¾!1­Â!1­Â!1=ç!1€=ç!1Û'"1üÛ'"1,"1,"1Q"1pQ"1­‘"1ì­‘"1Q–"1ôQ–"1yÚ"1`yÚ"1Ùþ"1ÌÙþ"10#1ì0#1ã~#1L ã~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s¾Š b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s¾Š b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s¾Š b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Oo  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj à àVVËË)@*@4µ4µ>(?(I›I›ST^�^�hôhôrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjààVVÌÌ B B ¸ ¸..¤¤��||òòh h"Þ"Þ$S%S'È'È)=*=,²,².'/'1œ1œ346†6†8û8û:p;p=å=å?Z@ZBÏBÏDDEDG¹G¹I,J,LŸLŸNOQ…Q…SøSøUkVkXÞXÞZQ[Q]Ä]Ä_7`7bªbªdeg�g�ijlvlvnénép\q\sÏsÏuBvBxµxµz({(}›}›€‚�‚�„ô„ô†g‡g‰Ú‰Ú‹MŒMŽÀŽÀ�3‘3“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1ï}!1�ï}!1�¾!1 �¾!1qÅ!1qÅ!1»ç!1€»ç!1Y("1üY("1C/"1C/"1�Q"1p�Q"1+’"1ì+’"1™"1ô™"1ÏØ"1`ÏØ"1/ý"1Ì/ý"10#1ì0#1ã~#1L ã~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1×€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1×€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1× €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥Å  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞSSÈÈ)=*=4²4²>%?%I˜I˜S T ^|^|hîhîr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽxxííb b"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†P‡P‰ÉË6Œ6ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1e}!1�e}!1¾!1 ¾!1ÑÂ!1ÑÂ!17ç!1€7ç!1Õ'"1üÕ'"1£,"1£,"1 Q"1p Q"1§‘"1ì§‘"1u–"1ôu–"1ÉØ"1`ÉØ"1)ý"1Ì)ý"10#1ì0#1Ý~#1L Ý~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1Õ€&1T Õ€&1¥R'1˜¥R'1Me'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r¿‹ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations17w€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r¿‹ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations17w€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r¿‹ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations17w €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥Å  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk å å[[ÑÑ)G*G4»4»>/?/I£I£ST^‹^‹hÿhÿrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkåå__ÙÙ S S Í ÍGGÁÁ;;µµ++¡¡ � �"#%y%y'ï'ï)e*e,Û,Û.Q/Q1Ç1Ç3=4=6³6³8)9);Ÿ;Ÿ=>@‡@‡BûBûDoEoGãGãIWJWLËLËN?O?Q³Q³S'T'V›V›XY[ƒ[ƒ]÷]÷_k`kbßbßdSeSgÇgÇi;j;l¯l¯n#o#q—q—s t vvxóxózg{g}Û}ÛO€O‚ÂÄ7…7‡©‡©‰ŠŒ�Œ�Ž�‘u‘u“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1õ}!1�õ}!1“¾!1 “¾!15Å!15Å!1Áç!1€Áç!1_("1ü_("1/"1/"1“Q"1p“Q"11’"1ì1’"1Ù˜"1ôÙ˜"1ÉØ"1`ÉØ"1)ý"1Ì)ý"10#1ì0#1Ý~#1L Ý~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1ç€&1T ç€&1·R'1˜·R'1_e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©J b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1›Û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©J b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1›Û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©J b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1›Û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥Å  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I¥I¥ST^Ž^Žhisrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLÍLÍNDODQ»Q»S2T2V©V©X Y [—[—]^`…`…bübüdsesgêgêiajalØlØnJoJq¼q¼s.t.v v xy{„{„}ö}öh€h‚Ú‚Ú„L…L‡¾‡¾‰0Š0Œ¢Œ¢Ž�‘†‘†“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1e}!1�e}!1¾!1 ¾!1ÑÂ!1ÑÂ!11ç!1€1ç!1Ï'"1üÏ'"1£,"1£,"1Q"1pQ"1¡‘"1ì¡‘"1u–"1ôu–"1ÉØ"1`ÉØ"1)ý"1Ì)ý"10#1ì0#1Ý~#1L Ý~#1Ÿ$1¬ Ÿ$1wò$1p wò$1±Þ%1t ±Þ%1]€&1T ]€&1-R'1˜-R'1Õd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p»‡ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations18x€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p»‡ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations18x€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p»‡ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations18x €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥Å  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áUUÉÉ)=*=4¯4¯>!?!I“I“ST^w^whéhérYsY} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‰‰ýýq q"å"å$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b�b�dÿdÿfqgqiãiãkUlUnÇnÇp9q9s«s«uvx�x�z{}s}såå�W‚W„ɄɆ9‡9‰ª‰ª‹ŒŽŒŽŒ�ý�ý’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1õ}!1�õ}!1“¾!1 “¾!15Å!15Å!1Áç!1€Áç!1_("1ü_("1/"1/"1“Q"1p“Q"11’"1ì1’"1Ù˜"1ôÙ˜"1ÉØ"1`ÉØ"1)ý"1Ì)ý"10#1ì0#1Ý~#1L Ý~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨J b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1›Û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨J b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1›Û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨J b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1›Û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥Å  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ù ÙNNÃÃ)8*84­4­>"?"I—I—S T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÙÙNNÃà 8 8 ­ ­""——  ��öökkààU U"Ê"Ê$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1_}!1�_}!1ý½!1 ý½!1¡Â!1¡Â!11ç!1€1ç!1Ï'"1üÏ'"1s,"1s,"1Q"1pQ"1¡‘"1ì¡‘"1E–"1ôE–"1ÉØ"1`ÉØ"1)ý"1Ì)ý"10#1ì0#1Ý~#1L Ý~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1�&1T �&1íR'1˜íR'1•e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¼ˆ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¼ˆ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¼ˆ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥Å  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg à àYYÒÒ)K*K4Ä4Ä>;?;I²I²S'T'^œ^œhis€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggààYYÒÒ K K Ä Ä==¶¶//¨¨!!šš Œ Œ"#%~%~'÷'÷)p*p,é,é.b/b1Û1Û3T4T6Í6Í8F9F;¿;¿=8>8@±@±B*C*E£E£GHJ“J“L M O�O�QøQøSoToVæVæX]Y][Ô[Ô]K^K`Â`Âb7c7e¬e¬g!h!j–j–l m o€o€qóqósftfvÙvÙxLyL{¿{¿}2~2€¥€¥‚ƒ…‹…‹‡ü‡ü‰nŠnŒàŒàŽR�R‘đē b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1õ}!1�õ}!1“¾!1 “¾!1kÅ!1kÅ!1Çç!1€Çç!1e("1üe("1=/"1=/"1™Q"1p™Q"17’"1ì7’"1™"1ô™"1yÚ"1`yÚ"1Ùþ"1ÔÙþ"10#1ô0#1Ý~#1T Ý~#1½$1´ ½$1åð$1x åð$1ÃÀ%1| ÃÀ%1ù€&1\ ù€&1ïP'1 ïP'1—c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²S b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²S b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²S b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Õ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1¬„Tx  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Þ ÞXXÎÎ)D*D4º4º>.?.I¢I¢ST^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÞÞXXÒÒ L L Æ Æ@@ºº44®®((¢¢ – –"#%†%†'ü'ü)r*r,è,è.^/^1Ô1Ô3J4J6À6À8696;¬;¬=">"@˜@˜BCE„E„GúGúInJnLâLâNVOVQÊQÊS>T>V²V²X&Y&[š[š]^`‚`‚böbödjejgÞgÞiRjRlÆlÆnDoDq¸q¸s,t,v v xy{ˆ{ˆ}ü}üp€p‚ä‚ä„X…X‡̇̉@Š@Œ´Œ´Ž(�(‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1_}!1�_}!1ý½!1 ý½!1¡Â!1¡Â!1+ç!1€+ç!1É'"1üÉ'"1s,"1s,"1ýP"1pýP"1›‘"1웑"1E–"1ôE–"1ÉØ"1`ÉØ"1)ý"1Ì)ý"10#1ì0#1Ý~#1L Ý~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¸„ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¸„ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¸„ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥Å  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>-?-I¢I¢ST^Š^Šhûhûrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbåbådXeXgËgËi>j>l±l±n&o&q—q—stvyvyxêxêz[{[}Ì}Ì=€=‚®‚®„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1õ}!1�õ}!1“¾!1 “¾!1eÅ!1eÅ!1Çç!1€Çç!1e("1üe("17/"17/"1™Q"1p™Q"17’"1ì7’"1 ™"1ô ™"1µÚ"1`µÚ"1ÿ"1Ôÿ"10#1ô0#1Ý~#1T Ý~#1¥$1´ ¥$1Íð$1x Íð$1«À%1| «À%1�€&1\ �€&1wP'1 wP'1c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±S b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ô€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±S b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ô€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±S b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ô €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1À˜‘µ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜTTÈÈ)<*<4°4°>"?"I”I”ST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœŒŒ | |"ô"ô$h%h'Ü'Ü)P*P,Ä,Ä.8/81¬1¬3 4 6”6”89;|;|=ð=ð?d@dBØBØDLELGÀGÀI2J2L¤L¤NOQˆQˆSúSúUlVlXÞXÞZP[P]Â]Â_4`4b¦b¦degŠgŠiüiüknlnnènèpZqZsÌsÌu>v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L%i!1Y}!1�Y}!1÷½!1 ÷½!1ÅÂ!1ÅÂ!1+ç!1€+ç!1É'"1üÉ'"1—,"1—,"1ýP"1pýP"1›‘"1웑"1i–"1ôi–"1sÚ"1`sÚ"1Óþ"1ÔÓþ"1 0#1ô 0#1×~#1T ×~#1Ï$1´ Ï$1÷ð$1x ÷ð$1ÕÀ%1| ÕÀ%15�&1\ 5�&1+Q'1 +Q'1Óc'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l¹… b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations14t€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l¹… b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations14t€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l¹… b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations14t €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1±‰Nr  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSÍÍ)E*E4½4½>3?3I©I©ST^¡^¡his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=>@•@•B C E…E…GýGýIsJsLéLéN_O_QÕQÕSKTKVÁVÁX7Y7[­[­]#^#`™`™bce…e…gûgûiqjqlçlçnioiqßqßsUtUvËvËxAyA{·{·}-~-€£€£‚ƒ…�…�‡ˆŠ{Š{ŒñŒñŽg�g‘ݑݓ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L%i!1û}!1�û}!1™¾!1 ™¾!1)Å!1)Å!1Íç!1€Íç!1k("1ük("1û."1û."1ŸQ"1pŸQ"1=’"1ì=’"1͘"1ô͘"1ŸØ"1`ŸØ"1ÿü"1Ìÿü"1 0#1ì 0#1×~#1L ×~#1«$1¬ «$1Óð$1p Óð$1±À%1t ±À%1{€&1T {€&1qP'1˜qP'1c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»\( b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1›Û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»\( b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1›Û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»\( b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1›Û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥¡  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10]] Ó ÓRRÇÇ)<*<4¯4¯>"?"I•I•ST^‡^‡húhúrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð]]ÓÓII¿¿ 5 5 « «!!——  ƒƒwwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Û=Û?N@NBÁBÁD4E4G§G§IJL�L�NOQsQsSæSæUYVYXÌXÌZ?[?]²]²_%`%b¤b¤degŠgŠiýiýkplpnãnãpVqVsÉsÉu'?'I›I›ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqƒqƒs÷s÷ukvkxßxßzS{S}Ç}Ç;€;‚¯‚¯„#…#‡—‡—‰ Š ŒŒŽóŽó�g‘g“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L%i!1û}!1�û}!1™¾!1 ™¾!1)Å!1)Å!1Íç!1€Íç!1k("1ük("1û."1û."1ŸQ"1pŸQ"1=’"1ì=’"1͘"1ô͘"1ŸØ"1`ŸØ"1ÿü"1Ìÿü"1 0#1ì 0#1×~#1L ×~#1™$1¬ ™$1Áð$1p Áð$1ŸÀ%1t ŸÀ%1€&1T €&1P'1˜P'1¹b'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†º\( b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1›Û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†º\( b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1›Û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†º\( b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1›Û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¥¡  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Õ ÕPPÃÃ)6*64§4§>?I‰I‰ST^w^whèhèrWsW} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……ssææY Y"Ì"Ì$?%?'²'²)%*%,˜,˜. / 1~1~3ñ3ñ5d6d8×8×:J;J=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éafbfd×d×fHgHi¹i¹k*l*n›n›p q s}s}uîuîw_x_zÐzÐ|A}A²²�#‚#„”„”†‡‰s‰s‹ã‹ã�SŽS�Ã�Ã’ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„i!15}!15}!1Ó½!1|Ó½!17Ã!1„7Ã!1mÆ!1ðmÆ!1÷Ï!1|÷Ï!1-Ó!1H-Ó!1ç!1Ôç!1Ÿ'"1PŸ'"1 -"1X -"1?0"1Ä?0"1É9"1PÉ9"1ÿ<"1ÿ<"1ÓP"1¨ÓP"1q‘"1$ q‘"1Û–"1, Û–"1š"1˜ š"1›£"1$ ›£"1Ѧ"1ð Ѧ"1Ô"1| Ô"1gÚ"1XgÚ"1Çþ"1ÔÇþ"1 ,#1ô ,#1é9#1Ðé9#1¥n#1\¥n#1'Œ#1€'Œ#1%—#1¬%—#1½$18½$1•ò$1ü•ò$1ÏÞ%1ÏÞ%1Iì%1àIì%1ù€&1lù€&1ÉR'1°ÉR'1qe'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PŸk b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1§ç bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PŸk b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1§ç  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PŸk  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1§ç bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¡Á§ÇRr bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations17w�Юî%eœÜ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d7·în¥%\Üïo“¤$ÊJYÙ�Ž8¸ïo¦&]Ý”ËK L ‚  9 ¹ ð p § ç  ^ • Õ  L  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10GG Û ÛUUÕÕ)K*K4Á4Á>7?7I«I«ST^Ÿ^Ÿhis‡s‡} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðGGÁÁ;;µµ / / à Ã==··11««%%ŸŸ “ “" # %�%�'(*y*y,ï,ï.e/e1Û1Û3Q4Q6Ç6Ç8=9=;³;³=)>)@Ÿ@ŸBCE‹E‹GHJwJwLëLëN_O_QÓQÓSGTGV»V»X/Y/[£[£]^`‹`‹bÿbÿdsesgçgçi[j[lÏlÏnOoOqÃqÃs7t7v«v«xy{“{“}~€{€{‚ï‚ï„c…c‡ׇ׉KŠKŒ¿Œ¿Ž3�3‘§‘§“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxi!1 ~!1è ~!1«¾!1d«¾!1MÅ!1lMÅ!1×Î!1è×Î!1 Ò!1´ Ò!1Ùç!1@Ùç!1w("1¼w("1/"1Ä/"1©8"1@©8"1ß;"1 ß;"1«Q"1˜«Q"1I’"1I’"1ñ˜"1 ñ˜"1{¢"1˜ {¢"1±¥"1d ±¥"1óÒ"1ð óÒ"1�Ú"1Ì �Ú"1ýþ"1P ýþ"1?,#1p?,#1[:#1L[:#1o#1Øo#1UŽ#1üUŽ#1S™#1(S™#1Ã$1´Ã$1›ò$1x›ò$1ÕÞ%1|ÕÞ%1Oì%1\Oì%1#�&1è#�&1óR'1,óR'1›e'10€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŸÓr> b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G‡±ñ¾þ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŸÓr> b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G‡±ñ¾þ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŸÓr>  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G‡±ñ¾þ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1¯Ï>^nj‰­ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1R’ä$I‰x¸ L  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dRÒ$¤övÈHšlì>¾�‹ ädVÖ¸8!¡Œ ìl ` à 4 ´  ˆ Ü \ ° 0 ‚  Tù 9¨è<| bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ý ÝVVÑÑ)J*J4©4©>@–@–BCEˆEˆGHJ”J”L M O‚O‚QùQùSpTpVçVçX^Y^[Õ[Õ]L^L`Ã`Ãb8c8e­e­g"h"j—j—l m o‹o‹qþqþsqtqväväxWyW{Ê{Ê}=~=€°€°‚#ƒ#…–…–‡ ˆ Š{Š{ŒíŒíŽ_�_‘ёѓ  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„i!1/}!1/}!1ͽ!1|ͽ!17Ã!1„7Ã!1mÆ!1ðmÆ!1÷Ï!1|÷Ï!1-Ó!1H-Ó!1ç!1Ôç!1Ÿ'"1PŸ'"1 -"1X -"1?0"1Ä?0"1É9"1PÉ9"1ÿ<"1ÿ<"1ÓP"1¨ÓP"1q‘"1$ q‘"1Û–"1, Û–"1š"1˜ š"1›£"1$ ›£"1Ѧ"1ð Ѧ"1Ô"1| Ô"1aÚ"1XaÚ"1Áþ"1ÔÁþ"1,#1ô,#1U:#1ÐU:#1o#1\o#11Ž#1€1Ž#1/™#1¬/™#1¥$18¥$1}ò$1ü}ò$1·Þ%1·Þ%11ì%1à1ì%1�€&1l�€&1QR'1°QR'1ùd'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oœh b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1§ç bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oœh b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1§ç  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oœh  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1§ç bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ž®§ÇLl bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1R’ä$Et´D  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dRÒ$¤övÈHšlì>¾ÃC�xøà`-­°0âb€ P Ð � Á    ð p À @ �  b ¢ ô 4„ÄT bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10FF Ù ÙQQÍÍ)A*A4µ4µ>)?)I›I›S T ^‹^‹hýhýroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðFF¾¾66®® & & ¹ ¹11©©!!™™‰‰ y y"ñ"ñ$m%m'á'á)U*U,É,É.=/=1±1±3%4%6™6™8 9 ;�;�=õ=õ?i@iBÝBÝDQEQGÅGÅI9J9L«L«NOQ�Q�STVsVsXåXåZW[W]É]É_;`;b­b­deg‘g‘ijlulunónópeqes×s×uIvIx»x»z-{-}Ÿ}Ÿ€‚ƒ‚ƒ„õ„õ†g‡g‰ىًKŒK޽޽�/‘/“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxi!1 ~!1è ~!1«¾!1d«¾!1MÅ!1lMÅ!1×Î!1è×Î!1 Ò!1´ Ò!1ßç!1@ßç!1}("1¼}("1/"1Ä/"1©8"1@©8"1ß;"1 ß;"1±Q"1˜±Q"1O’"1O’"1ñ˜"1 ñ˜"1{¢"1˜ {¢"1±¥"1d ±¥"1óÒ"1ð óÒ"1+Ú"1Ì +Ú"1‹þ"1P ‹þ"1Í+#1pÍ+#1:#1L:#1;o#1Ø;o#1�#1ü�#1 š#1( š#1±$1´±$1‰ò$1x‰ò$1ÃÞ%1|ÃÞ%1=ì%1\=ì%1±€&1è±€&1�R'1,�R'1)e'10€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ôt@ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Hˆ²ò¿ÿ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ôt@ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Hˆ²ò¿ÿ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ôt@  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Hˆ²ò¿ÿ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1–’/±‰Ë7 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1]�ú:•ÕÔ 0p  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d]Ý:º—ôtÑQ®.‹ hèCÞùy Ô T · 7 ‚  ¯ / M Í Š  ˜ e å @ À ]„ļüW—ò2 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPËË)B*B4ž4ž>0?0I¥I¥ST^•^•hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=ê=ê?a@aBØBØDOEOGÆGÆIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n1o1q¢q¢stv„v„xõxõzf{f}×}×H€H‚¹‚¹„*…*‡œ‡œ‰ Š Œ~Œ~ŽïŽï�`‘`“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1Y}!1èY}!1÷½!1d÷½!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!1%ç!1@%ç!1Ã'"1¼Ã'"1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1÷P"1˜÷P"1•‘"1•‘"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"1aÚ"1Ì aÚ"1Áþ"1H Áþ"1,#1h,#1Ç:#1DÇ:#1ƒo#1Ѓo#1e�#1ôe�#1c›#1 c›#1Ÿ$1¬Ÿ$1wò$1pwò$1±Þ%1t±Þ%1+ì%1T+ì%1u€&1àu€&1ER'1$ER'1íd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations17kº† b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @€7wÈ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations17kº† b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @€7wÈ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations17kº†  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @€7wÈ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ž¾7Ll bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1l¬XBÆt´  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dlìXØDÄ0°œˆôtÌLƒà`:ºÎNñq¼< ª * ˜  †  t ô r ² b â PÐ<|è(–ÖD„ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ Ô ÔII¾¾)3*34Ž4Ž>?IŽIŽS T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÔÔJJÀÀ 6 6 ¬ ¬""˜˜„„ùùnnããX X"Í"Í$B%B'·'·),*,,¡,¡./1‹1‹346u6u8ê8ê:_;_=º=º?-@-B B DEG†G†IJL†L†NùNùPlQlSßSßURVRXÅXÅZ8[8]«]«_`b›b›deg�g�iôiôkglgnÚnÚpMqMsÀsÀu3v3x¦x¦z{}Œ}Œÿÿ�r‚r„å„å†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1¥¾!1d¥¾!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!1Óç!1@Óç!1q("1¼q("1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1¥Q"1˜¥Q"1C’"1C’"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"1‘Ú"1Ì ‘Ú"1ñþ"1P ñþ"13,#1p3,#1‘7#1L‘7#1;{#1È;{#1íŒ#1ìíŒ#1ë—#1Èë—#1Ÿ$1TŸ$1wò$1wò$1±Þ%1±Þ%1+ì%1ü+ì%1 �&1ˆ �&1ÛR'1ÌÛR'1ƒe'1Ѐ} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”Èg3 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R’™ÙÉ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”Èg3 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R’™ÙÉ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”Èg3  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R’™ÙÉ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1¨È]}`\t˜ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 I‰Ò@€ bÒ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1<IÉ’Û[¤$mí6¶ÿÈHvö‘ÑZ?¿ˆÑQ Ð  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜTTÌÌ)D*D4¼4¼>4?4IªIªS T ^˜^˜hisjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœŒŒ | |"ô"ô$l%l'ä'ä)\*\,Ô,Ô.L/L1Ä1Ä3<4<6´6´8,9,;¤;¤=>@”@”B C E„E„GüGüItJtLêLêN`O`QÖQÖSLTLVÂVÂX8Y8[®[®]$^$`š`šbce†e†gügüirjrlèlèn`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡â‡â‰XŠXŒÎŒÎŽD�D‘º‘º“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1S}!1èS}!1ñ½!1dñ½!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!1%ç!1@%ç!1Ã'"1¼Ã'"1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1÷P"1˜÷P"1•‘"1•‘"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"1mÚ"1Ì mÚ"1Íþ"1H Íþ"1,#1h,#1å:#1Då:#1¡o#1Сo#1ï�#1ôï�#1í›#1 í›#1�$1¬�$1eò$1peò$1ŸÞ%1tŸÞ%1ì%1Tì%1€&1à€&1ßQ'1$ßQ'1‡d'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations17k¸„ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ?7wÇ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations17k¸„ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ?7wÇ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations17k¸„  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ?7wÇ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‘±èTt bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1uµ*jÝy¹ �Ð  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1duõjê_ßTÔIÉ>¾3³(¨›¨(  Ž _ ß  �  – ô t Í M ç g Ú Z é ) Í MÀ@µõjª]Ð bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ö ÖII¼¼)/*/4‡4‡>?I„I„SýSý]n^nhßhßrPsP} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††ùùllßßR R"Å"Å$8%8'«'«)*,‘,‘./1w1w3ê3ê5]6]8Ð8Ð:C;C=›=›? @ B}B}DîDîF_G_IëIëK\L\NÍNÍP>Q>S¯S¯U V X‘X‘Z[]s]s_ä_äa]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ü†üˆl‰l‹Ü‹Ü�LŽL�¼�¼’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1¥¾!1d¥¾!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!1Ùç!1@Ùç!1w("1¼w("1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1«Q"1˜«Q"1I’"1I’"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"1OÚ"1Ì OÚ"1¯þ"1P ¯þ"1ñ+#1pñ+#1:#1L:#1Ã}#1ÈÃ}#1q�#1ìq�#1o˜#1Èo˜#1�$1T�$1eò$1eò$1ŸÞ%1ŸÞ%1ì%1üì%1“€&1ˆ“€&1cR'1ÌcR'1 e'1Ѐ} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•Éi5 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“™ÙÊ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•Éi5 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“™ÙÊ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•Éi5  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“™ÙÊ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1�‹2R£{»' bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 7w®òÉ  bÒ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1<7·în¥%\Ü“ÊJ�8¸ï/f¦š  c ã , ¬ õ u Y™ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÄÄ):*:4°4°>&?&IšIšST^†^†húhúrSsS} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$L%L'Â'Â)8*8,®,®.$/$1š1š346†6†8ü8ü:r;r=è=è?^@^BÔBÔDJEJGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡s‡s‰ç‰ç‹[Œ[ŽÏŽÏ�C‘C“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1_}!1è_}!1ý½!1dý½!1kÅ!1lkÅ!1õÎ!1èõÎ!1+Ò!1´+Ò!1+ç!1@+ç!1É'"1¼É'"1=/"1Ä=/"1Ç8"1@Ç8"1ý;"1 ý;"1ýP"1˜ýP"1›‘"1›‘"1™"1 ™"1™¢"1˜ ™¢"1Ï¥"1d Ï¥"1Ó"1ð Ó"1UÚ"1Ì UÚ"1µþ"1H µþ"1÷+#1h÷+#1O7#1DO7#1ùz#1Àùz#1¯‹#1䯋#1­–#1À­–#1�$1L�$1eò$1eò$1ŸÞ%1ŸÞ%1ì%1ôì%1“€&1€“€&1cR'1ÄcR'1 e'1È€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l»‡ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T7wÓ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l»‡ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T7wÓ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l»‡  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T7wÓ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‘±6V2R bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 c£Jý= bÒ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1<cãFÆ)© ŒïoÒRµ5ÓS˜Œ {»FÆ^ÿŽ Î  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ù ÙNNÃÃ)8*84­4­>#?#I—I—S T ^^hóhórOsO} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A ¹ ¹11©©!!™™ƒƒøøm m"â"â$W%W'Ì'Ì)A*A,¶,¶.+/+1 1 346Š6Š8ÿ8ÿ:t;t=é=é?^@^BÓBÓDHEHG½G½I3J3L§L§NOQ�Q�STVwVwXëXëZ_[_]Ó]Ó_G`Gb»b»d/e/g£g£ijl‹l‹nÿnÿpsqssçsçu[v[xÏxÏzC{C}·}·+€+‚Ÿ‚Ÿ„…‡o‡o‰ä‰ä‹YŒYŽÎŽÎ�C‘C“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l%i!1û}!1Ðû}!1™¾!1L™¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1Íç!1(Íç!1k("1¤k("1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1ŸQ"1€ŸQ"1=’"1ü=’"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1yÚ"1´ yÚ"1Ùþ"18 Ùþ"1,#1X,#17#147#1½Œ#1 ½Œ#1·$1,·$1�ò$1ð�ò$1ÉÞ%1ôÉÞ%1í€&1Ôí€&1½R'1½R'1ee'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½\( b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]�ØÖ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½\( b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]�ØÖ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½\(  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]�ØÖ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1�½XxJFNr bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1/o€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1YÙ  çg®. Ÿ ß  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I£I£ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbèbèd]e]gÒgÒiGjGl¼l¼n1o1q¦q¦stv�v�xy{z{z}ï}ïd€d‚قلN…N‡ÇÉ8Š8Œ­Œ­Ž"�"‘—‘—“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1Y}!1èY}!1÷½!1d÷½!1eÅ!1leÅ!1ïÎ!1èïÎ!1%Ò!1´%Ò!1+ç!1@+ç!1É'"1¼É'"17/"1Ä7/"1Á8"1@Á8"1÷;"1 ÷;"1ýP"1˜ýP"1›‘"1›‘"1 ™"1  ™"1“¢"1˜ “¢"1É¥"1d É¥"1 Ó"1ð  Ó"1aÚ"1Ì aÚ"1Áþ"1H Áþ"1,#1h,#1Ë9#1DË9#1u}#1Àu}#1'Œ#1ä'Œ#1%—#1À%—#1{$1L{$1Sò$1Sò$1�Þ%1�Þ%1ì%1ôì%1Q€&1€Q€&1!R'1Ä!R'1Éd'1È€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l¹… b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U7wÒ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l¹… b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U7wÒ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l¹…  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U7wÒ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Šª +Dd bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 <|¸üv¶  bÒ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1<<¼øx´4pð,¬èh¤$`à\˜Ø¸8 q ñ * ª ã c  G  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×LLÁÁ)6*64«4«> ? I“I“ST^w^whéhérBsB} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡üüqqææ[ ["Ð"Ð$E%E'º'º)/*/,¤,¤./1Ž1Ž346x6x8í8í:b;b=×=×?L@LBÁBÁD6E6G«G«I J L“L“NOQyQySìSìU_V_XÒXÒZE[E]¸]¸_+`+b�b�deg�g�ióiókelen×n×pIqIs»s»u-v-xŸxŸz{}ƒ}ƒõõ�g‚g„لن2‡2‰¥‰¥‹Œދދ�þ�þ’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l%i!1~!1Ð~!1Ÿ¾!1LŸ¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Íç!1(Íç!1k("1¤k("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1ŸQ"1€ŸQ"1=’"1ü=’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1aÚ"1´ aÚ"1Áþ"18 Áþ"1,#1X,#1:#14:#1G�#1 G�#1¥$1,¥$1}ò$1ð}ò$1·Þ%1ô·Þ%1‡€&1Ô‡€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1о^* b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ^žÚ× bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1о^* b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ^žÚ× bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1о^*  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ^žÚ×  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1ˆ„ŒdCc« bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¶ö €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1u õ < ¼  ƒ Ê J &f €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙOOÄÄ)9*94®4®>#?#I–I–S T ^|^|hïhïrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰ÿÿuuëëa a"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnänäpWqWsÊsÊu=v=x°x°z#{#}–}– € ‚|‚|„ï„ï†b‡b‰Õ‰Õ‹HŒH޻޻�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1_}!1Ð_}!1ý½!1Lý½!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!11ç!1(1ç!1Ï'"1¤Ï'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1Q"1€Q"1¡‘"1ü¡‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1O7#1,O7#1Í‹#1˜Í‹#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1á€&1Ìá€&1±R'1±R'1Ye'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¼ˆ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 5u9yà  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¼ˆ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 5u9yà  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¼ˆ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 5u9yà   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 „¤8Ss bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ø8€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÚZ•PÐ ‹¨ è  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>-?-I¤I¤ST^�^�hÿhÿrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜIUJULÌLÌNCOCQºQºS1T1V¨V¨XY[–[–] ^ `„`„bûbûdrergégéi`j`l×l×nIoIq»q»s-t-vŸvŸxy{ƒ{ƒ}õ}õg€g‚قلK…K‡½‡½‰/Š/Œ¡Œ¡Ž�‘…‘…“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1õ}!1Ðõ}!1“¾!1L“¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Çç!1(Çç!1e("1¤e("1[/"1¬[/"1å8"1(å8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1UÚ"1´ UÚ"1µþ"18 µþ"1÷+#1X÷+#1s7#14s7#1“Œ#1 “Œ#1·$1,·$1�ò$1ð�ò$1ÉÞ%1ôÉÞ%1Ï€&1ÔÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²Q b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 h¨™Ùî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²Q b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 h¨™Ùî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²Q  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 h¨™Ùî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Ç’²401U bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1^€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1<¼�ÆF‹ n ®  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh â âXXÎÎ)D*D4º4º>.?.I¢I¢ST^Š^Šhþhþrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²((žž Š Š"#%v%v'ì'ì)b*b,Ø,Ø.N/N1Ä1Ä3:4:6°6°8&9&;œ;œ=>@†@†BúBúDnEnGâGâIVJVLÊLÊN>O>Q²Q²S&T&VšVšXY[‚[‚]ö]ö_j`jbÞbÞdReRgÆgÆi:j:l®l®n"o"q–q–s t v~v~xòxòzf{f}Ú}ÚN€N‚‚„6…6‡ª‡ª‰ŠŒ�Œ�Ž�‘v‘v“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1_}!1Ð_}!1ý½!1Lý½!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!11ç!1(1ç!1Ï'"1¤Ï'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1Q"1€Q"1¡‘"1ü¡‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1Ë9#1,Ë9#1EŒ#1˜EŒ#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19mº† b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 4t8xÞ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19mº† b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 4t8xÞ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19mº†  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 4t8xÞ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �¡0P2R bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1s³ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÆF �  < ¼ ÷ w # c  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ö ÖKKÀÀ)5*54ª4ª>!?!I–I–S T ^}^}hïhïrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÖÖKKÀÀ 5 5 ª ª””  ~~óóhhÝÝR R"Ç"Ç$<%<'±'±)&*&,›,›./1…1…3ú3ú5o6o8ä8ä:Y;Y=Î=Î?C@CB¸B¸D-E-G¢G¢IJLŽLŽNOQxQxSíSíUbVbX×X×ZL[L]Á]Á_6`6b«b«d e g•g•i j llnñnñpcqcsÕsÕuGvGx¹x¹z+{+}�}�€‚�‚�„ó„ó†e‡e‰׉׋IŒI޻޻�-‘-“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1û}!1Ðû}!1™¾!1L™¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Çç!1(Çç!1e("1¤e("1[/"1¬[/"1å8"1(å8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1Ú"1´ Ú"1gþ"18 gþ"1©+#1X©+#1û9#14û9#1�#1 �#1Ÿ$1,Ÿ$1wò$1ðwò$1±Þ%1ô±Þ%1W€&1ÔW€&1'R'1'R'1Ïd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³S b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 i©™Ùð0 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³S b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 i©™Ùð0 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³S  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 i©™Ùð0  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Ç�}~V› bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1£ã €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1P Ð  • Ú Z Ÿ  ó 3 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞRRÆÆ):*:4®4®> ? I’I’ST^v^vhèhèrZsZ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žž††úún n"â"â$V%V'Ê'Ê)>*>,²,².&/&1š1š346‚6‚8ö8ö:j;j=Þ=Þ?P@PBÂBÂD4E4G¦G¦IJLŠLŠNüNüPnQnSàSàURVRXÄXÄZ6[6]¨]¨_`bŒbŒdþdþfpgpiâiâkTlTnÆnÆp8q8sªsªuvxŽxŽz{}r}rää�V‚V„ȄȆ:‡:‰«‰«‹ŒŽ�Ž��þ�þ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1e}!1Ðe}!1¾!1L¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!17ç!1(7ç!1Õ'"1¤Õ'"1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1 Q"1€ Q"1§‘"1ü§‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1a7#1,a7#1]Œ#1˜]Œ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1ó€&1Ìó€&1ÃR'1ÃR'1ke'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n½‰ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IO�@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n½‰ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IO�@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n½‰  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IO�@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w—þPp bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�ÔT˜\Ü4 t  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee ß ßYYÏÏ)E*E4»4»>1?1I¥I¥ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯))££ — —"#%‡%‡'ý'ý)s*s,é,é._/_1Õ1Õ3K4K6Á6Á8797;­;­=#>#@™@™BCE…E…GûGûIqJqLåLåNYOYQÍQÍSATAVµVµX)Y)[�[�]^`…`…bùbùdmemgágáiUjUlÉlÉn=o=q±q±s%t%v™v™x y {�{�}õ}õi€i‚݂݄Q…Q‡Ňʼn9Š9Œ­Œ­Ž!�!‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ï}!1Ðï}!1�¾!1L�¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Áç!1(Áç!1_("1¤_("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1CÚ"1´ CÚ"1£þ"18 £þ"1å+#1Xå+#1g7#14g7#1cŒ#1 cŒ#1É$1,É$1¡ò$1ð¡ò$1ÛÞ%1ôÛÞ%1)�&1Ô)�&1ùR'1ùR'1¡e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§F b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 h¨Ê â" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§F b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 h¨Ê â" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§F  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 h¨Ê â"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Ç‡§> bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 M€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ÿâb¥%hè= }  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh á áZZÓÓ)L*L4Å4Å>>?>IµIµS*T*^�^�hisƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhááZZÓÓ L L Å Å>>··00©©""›› � �"#%%'ø'ø)q*q,ê,ê.c/c1Ü1Ü3U4U6Î6Î8G9G;À;À=9>9@²@²B+C+E¤E¤GHJ–J–L M O„O„QûQûSrTrVéVéX`Y`[×[×]N^N`Å`Åb:c:e¯e¯g$h$j™j™lmo�o�qôqôsgtgvÚvÚxMyM{À{À}3~3€¦€¦‚ƒ…Œ…Œ‡ÿ‡ÿ‰qŠqŒãŒãŽU�U‘ǑǓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1e}!1Ðe}!1¾!1L¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!17ç!1(7ç!1Õ'"1¤Õ'"1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1 Q"1€ Q"1§‘"1ü§‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1ã9#1,ã9#1áŒ#1˜áŒ#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n»‡ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JNŽþ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n»‡ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JNŽþ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n»‡  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JNŽþ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x˜?_ @ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ˆÈ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1   ä d ¨ ( l ì ¸ ø  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝUUÉÉ)=*=4±4±>%?%I—I—S T ^{^{híhír_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%���� } }"õ"õ$i%i'Ý'Ý)Q*Q,Å,Å.9/91­1­3!4!6•6•8 9 ;};}=ñ=ñ?e@eBÙBÙDMEMGÁGÁI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnãnãpUqUsÇsÇu9v9x«x«z{}�}�€‚s‚s„å„å†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ï}!1Ðï}!1�¾!1L�¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1Áç!1(Áç!1_("1¤_("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1ûÙ"1´ ûÙ"1[þ"18 [þ"1�+#1X�+#1é9#14é9#1çŒ#1 çŒ#1·$1,·$1�ò$1ð�ò$1ÉÞ%1ôÉÞ%1±€&1Ô±€&1�R'1�R'1)e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨H b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 i©Ë ä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨H b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 i©Ë ä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨H  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 i©Ë ä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1§ÇzvvN‹÷ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Ð €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1+ « î n ± 1 t ô À  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ý ÝTTËË)B*B4¹4¹>0?0I¥I¥ST^‰^‰húhúrlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÝÝTTËË B B ¹ ¹00§§••  ƒƒúúq q"è"è$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n%o%q–q–stvxvxxéxézZ{Z}Ë}Ë<€<‚­‚­„…‡�‡�‰ŠŒrŒrŽãŽã�T‘T“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1k}!1Ðk}!1 ¾!1L ¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1=ç!1(=ç!1Û'"1¤Û'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1Q"1€Q"1­‘"1ü­‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1 Ú"1´  Ú"1mþ"10 mþ"1¯+#1P¯+#1y7#1,y7#1#�#1˜#�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1{€&1Ì{€&1KR'1KR'1ód'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¾Š b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 *j7wÉ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¾Š b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 *j7wÉ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¾Š  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 *j7wÉ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 jЧÇä bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Z€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hè7·†ÕU ú : €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØMMÂÂ)7*74¬4¬>?I’I’ST^x^xhëhër^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?K@KB¾B¾D1E1G¤G¤IJLŠLŠNýNýPpQpSãSãUVVVXÉXÉZ<[<]¯]¯_"`"b•b•deg{g{iîiîkalanÔnÔpGqGsºsºu-v-x x z{}†}†ùù�l‚l„߄߆R‡R‰ĉċ6Œ6ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1é}!1Ðé}!1‡¾!1L‡¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1»ç!1(»ç!1Y("1¤Y("1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1�Q"1€�Q"1+’"1ü+’"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"11Ú"1´ 1Ú"1‘þ"18 ‘þ"1Ó+#1XÓ+#1s7#14s7#1½Œ#1 ½Œ#1Õ$1,Õ$1­ò$1ð­ò$1çÞ%1ôçÞ%1;�&1Ô;�&1 S'1 S'1³e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœ; b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘È�Ý bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœ; b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘È�Ý bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœ;  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘È�Ý  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Ç|œ( bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1R€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1HÈ�ÖV� ’ Ò  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àXXÐÐ)H*H4À4À>8?8I®I®S$T$^š^šhis†s†} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  �� € €"ø"ø$p%p'è'è)`*`,Ø,Ø.P/P1È1È3@4@6¸6¸8090;¨;¨= > @˜@˜BCEˆEˆGHJxJxLîLîNdOdQÚQÚSPTPVÆVÆX?IˆIˆSùSù]j^jhÛhÛrLsL} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠýýppããV V"É"É$<%<'¯'¯)"*",•,•./1{1{3î3î5a6a8Ô8Ô:G;G=º=º?+@+BœBœD E G~G~IïIïK`L`NÑNÑPBQBS³S³U$V$X•X•Z[]w]w_è_èaYbYdÊdÊf;g;i¬i¬klnŽnŽpÿpÿrpspuáuáwRxRzÃzÃ|4}4¥¥�‚„‡„‡†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1é}!1Ðé}!1‡¾!1L‡¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1»ç!1(»ç!1Y("1¤Y("1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1�Q"1€�Q"1+’"1ü+’"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1ïÙ"1´ ïÙ"1Oþ"18 Oþ"1‘+#1X‘+#1õ9#14õ9#1A�#1 A�#1½$1,½$1•ò$1ð•ò$1ÏÞ%1ôÏÞ%1À&1ÔÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 P�ÇžÞ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 P�ÇžÞ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 P�ÇžÞ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1§ÇsooG{ç bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1™Ù €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1d ä + « ò r ¹ 9 Y €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜRRÈÈ)>*>4´4´>*?*IžIžST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒxxîîd d"Ú"Ú$P%P'Æ'Æ)<*<,²,².(/(1ž1ž346Š6Š89;v;v=ì=ì?b@bBØBØDNENGÄGÄI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1q}!1Ðq}!1¾!1L¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Cç!1(Cç!1á'"1¤á'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1Q"1€Q"1³‘"1ü³‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1ûÙ"1´ ûÙ"1[þ"10 [þ"1�+#1P�+#1‘7#1,‘7#1ï�#1˜ï�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p¿‹ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @8xÍ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p¿‹ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @8xÍ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p¿‹  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @8xÍ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 §Ç]}Óó bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations10p€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1À@štôN Î À €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝRRÇÇ)<*<4±4±>'?'I›I›ST^ƒ^ƒh÷h÷rmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��‡‡üüq q"æ"æ$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡�‡�‰ŠŒwŒwŽìŽì�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1ã}!1Ðã}!1�¾!1L�¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1¯ç!1(¯ç!1M("1¤M("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1�Q"1€�Q"1’"1ü’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1Ú"1´ Ú"1þ"18 þ"1Á+#1XÁ+#1‹7#14‹7#1‰�#1 ‰�#1É$1,É$1¡ò$1ð¡ò$1ÛÞ%1ôÛÞ%1�&1Ô�&1ÕR'1ÕR'1}e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘0ü b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T”Ë Á bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘0ü b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T”Ë Á bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘0ü  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T”Ë Á  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Çr’óïï bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1(h€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  ròDÄ – X˜ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>2?2I§I§ST^‘^‘his{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbìbìdaeagÖgÖiKjKlÀlÀn5o5qªqªstv”v”x y {~{~}ó}óh€h‚݂݄R…R‡LJlj<Š<Œ±Œ±Ž&�&‘›‘›“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1q}!1Ðq}!1¾!1L¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1=ç!1(=ç!1Û'"1¤Û'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1Q"1€Q"1­‘"1ü­‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1%Ú"1´ %Ú"1…þ"10 …þ"1Ç+#1PÇ+#1+:#1,+:#1‹Ž#1˜‹Ž#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p½‰ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @8xÎ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p½‰ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @8xÎ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p½‰  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @8xÎ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 §Çf†ü bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ê €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1( ¨  ‚ Ü \ ¶ 6 Zš €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛPPÅÅ):*:4¯4¯>$?$I—I—S T ^{^{híhírasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹uuêê_ _"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I$J$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b¡b¡deg…g…i÷i÷kilinÛnÛpMqMs¿s¿u1v1x£x£z{}‡}‡ùù�k‚k„݄݆Q‡Q‰ĉċ7Œ7ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1ã}!1Ðã}!1�¾!1L�¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1µç!1(µç!1S("1¤S("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1‡Q"1€‡Q"1%’"1ü%’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1ãÙ"1´ ãÙ"1Cþ"18 Cþ"1…+#1X…+#1:#14:#1Ž#1 Ž#1·$1,·$1�ò$1ð�ò$1ÉÞ%1ôÉÞ%1™€&1Ô™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’2þ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“Ê Á bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’2þ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“Ê Á bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’2þ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“Ê Á  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Çlhh@k× bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ºú €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1è h º : Œ ^ Þ ê* €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝSSÈÈ)=*=4²4²>'?'IšIšS T ^€^€hóhórfsf} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��yyïïe e"Û"Û$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I+J+LžLžNOQ„Q„S÷S÷UjVjXÝXÝZP[P]Ã]Ã_6`6b©b©deg�g�ijlulunènèp[q[sÎsÎuAvAx´x´z'{'}š}š € ‚€‚€„ó„ó†f‡f‰ىًLŒLŽ¿Ž¿�2‘2“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1w}!1Ðw}!1¾!1L¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Cç!1(Cç!1á'"1¤á'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1Q"1€Q"1³‘"1ü³‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1éÙ"1´ éÙ"1Iþ"10 Iþ"1‹+#1P‹+#1¯7#1,¯7#1»Ž#1˜»Ž#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ó€&1Ìó€&1ÃR'1ÃR'1ke'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÀŒ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _7wÏ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÀŒ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _7wÏ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÀŒ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _7wÏ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 §ÇPp½Ý bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1F†€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1˜ý}âb Ç G †Æ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>1?1I¨I¨ST^‘^‘hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIYJYLÐLÐNGOGQ¾Q¾S5T5V¬V¬X#Y#[š[š]^`ˆ`ˆbÿbÿdvevgígíidjdlÛlÛnMoMq¿q¿s1t1v£v£xy{‡{‡}ù}ùk€k‚݂݄O…O‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1Ý}!1ÐÝ}!1{¾!1L{¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1©ç!1(©ç!1G("1¤G("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1{Q"1€{Q"1’"1ü’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1Ú"1´ Ú"1ßþ"18 ßþ"1!,#1X!,#1£7#14£7#1OŽ#1 OŽ#1É$1,É$1¡ò$1ð¡ò$1ÛÞ%1ôÛÞ%1ç€&1Ôç€&1·R'1·R'1_e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†%ñ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—Ψè bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†%ñ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—Ψè bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†%ñ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—묏  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1f†ÜØKkØü bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1>~€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1øxÕU²2 �  ^ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll æ æ\\ÒÒ)H*H4¾4¾>2?2I¦I¦ST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllææ``ÚÚ T T Î ÎHHÂÂ<<¶¶,,¢¢ Ž Ž"#%z%z'ð'ð)f*f,Ü,Ü.R/R1È1È3>4>6´6´8*9*; ; =>@Š@ŠBþBþDrErGæGæIZJZLÎLÎNBOBQ¶Q¶S*T*VžVžXY[†[†]ú]ú_n`nbâbâdVeVgÊgÊi>j>l²l²n&o&qšqšstv‚v‚xöxözj{j}Þ}ÞR€R‚ƂƄ:…:‡®‡®‰!Š!Œ”Œ”Ž�‘z‘z“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1q}!1Ðq}!1¾!1L¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Cç!1(Cç!1á'"1¤á'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1Q"1€Q"1³‘"1ü³‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1Ú"1´ Ú"1yþ"10 yþ"1»+#1P»+#1O:#1,O:#1]�#1˜]�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1Ÿ€&1ÌŸ€&1oR'1oR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q¾Š b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ^7wÐ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q¾Š b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ^7wÐ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=q¾Š  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ^7wÐ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 §Ç]}ê  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ë+ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¬ , ‘  v ö [ Û +k €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ú ÚOOÄÄ)9*94®4®>%?%IšIšST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÚÚOOÄÄ 9 9 ® ®##˜˜  ‚‚÷÷llááV V"Ë"Ë$@%@'µ'µ)***,Ÿ,Ÿ./1‰1‰3þ3þ5s6s8è8è:];]=Ò=Ò?G@GB¼B¼D1E1G¦G¦IJL’L’NOQ|Q|SñSñUfVfXÛXÛZP[P]Å]Å_:`:b¯b¯d$e$g™g™ijlƒlƒnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1Ý}!1ÐÝ}!1{¾!1L{¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1¯ç!1(¯ç!1M("1¤M("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1�Q"1€�Q"1’"1ü’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1…Ú"1´ …Ú"1åþ"18 åþ"1',#1X',#1C:#14C:#1ëŽ#1 ëŽ#1±$1,±$1‰ò$1ð‰ò$1ÃÞ%1ôÃÞ%1o€&1Ôo€&1?R'1?R'1çd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡'ó b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V–Í ©é bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡'ó b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V–Í ©é bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡'ó  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V–Í ©é  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1eaa9Rr[Ç bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Û €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1l ì I É & ¦  ƒ »û €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âVVÊÊ)>*>4²4²>$?$I–I–ST^z^zhìhìr^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢ŠŠþþr r"æ"æ$Z%Z'Î'Î)B*B,¶,¶.*/*1ž1ž346†6†8ú8ú:n;n=â=â?T@TBÆBÆD8E8GªGªIJLŽLŽNOQrQrSäSäUVVVXÈXÈZ:[:]¬]¬_`b�b�degtgtiæiækXlXnÊnÊp‡>‰¯‰¯‹ Œ ޑޑ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1ã}!1�ã}!1�¾!1�¾!1­Â!1$­Â!1µç!1�µç!1S("1S("1,"1$,"1‡Q"1�‡Q"1%’"1%’"1Q–"1$Q–"1áØ"1�áØ"1Aý"1üAý"1)0#1 )0#1õ~#1| õ~#1ç$1Ü ç$1¿ò$1  ¿ò$1ùÞ%1¤ ùÞ%1�&1„ �&1íR'1ÈíR'1•e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuOƒÒž-ù b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuOƒÒž-ù b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuOƒÒž-ù b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Ç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10zz ô ônnää)Z*Z4Ð4Ð>D?DI¸I¸S,T,^ ^ hisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðzzôônnèè b b Ü ÜVVÐÐJJÄÄ>>¸¸22 ¬ ¬"&#&%œ%œ'(*ˆ*ˆ,þ,þ.t/t1ê1ê3`4`6Ö6Ö8L9L;Â;Â=8>8@®@®B$C$EšEšGHJ„J„LøLøNlOlQàQàSTTTVÈVÈXI?IIÀIÀS5T5^¨^¨hisŒsŒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuîîggàà Y Y Ò ÒKKÄÄ==¶¶//¨¨!! š š"#%Œ%Œ'(*~*~,÷,÷.p/p1é1é3b4b6Û6Û8T9T;Í;Í=F>F@¿@¿B8C8E±E±G*H*J¡J¡LMO�O�QRT}T}VôVôXkYk[â[â]Y^Y`Ð`ÐbEcEeºeºg/h/j¤j¤lmoŒoŒqÿqÿsrtrvåvåxXyX{Ë{Ë}>~>€±€±‚$ƒ$…—…—‡ˆŠzŠzŒìŒìŽ^�^‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1é}!1�é}!1‡¾!1‡¾!1­Â!1$­Â!1»ç!1�»ç!1Y("1Y("1,"1$,"1�Q"1��Q"1+’"1+’"1Q–"1$Q–"1áØ"1�áØ"1Aý"1üAý"1)0#1 )0#1õ~#1| õ~#1Ï$1Ü Ï$1§ò$1  §ò$1áÞ%1¤ áÞ%1¥€&1„ ¥€&1uR'1ÈuR'1e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvM�Κ0ü b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvM�Κ0ü b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvM�Κ0ü b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Ç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx ð ðhhÜÜ)P*P4Ä4Ä>6?6I¨I¨ST^Œ^Œhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðxxððhhàà X X Ð ÐHHÀÀ88°°((   � �"#%|%|'ð'ð)d*d,Ø,Ø.L/L1À1À34446¨6¨89;�;�=>@x@xBìBìD`E`GÔGÔIFJFL¸L¸N*O*QœQœSTV€V€XòXòZd[d]Ö]Ö_H`Hbºbºd,e,gžgžijl‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1×}!1�×}!1u¾!1 u¾!1wÅ!1wÅ!1©ç!1€©ç!1G("1üG("1I/"1I/"1{Q"1p{Q"1’"1ì’"1™"1ô™"1‹Ú"1`‹Ú"1ëþ"1Ìëþ"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1Õ€&1T Õ€&1¥R'1˜¥R'1Me'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Eyå b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Í €} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Eyå b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Í €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Eyå b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Í  €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Nn  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu ì ìccÚÚ)Q*Q4È4È>=?=I²I²S%T%^–^–hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuììccÚÚ Q Q È È??¶¶--¤¤’’  € €"÷"÷$n%n'å'å)\*\,Ó,Ó.J/J1Á1Á38486¯6¯8&9&;�;�=>@‹@‹BCEyEyGðGðIeJeLÚLÚNOOOQÄQÄS9T9V®V®X#Y#[˜[˜] ^ `‚`‚bõbõdhehgÛgÛiNjNlÁlÁn2o2q£q£stv…v…xöxözg{g}Ø}ØI€I‚º‚º„+…+‡›‡›‰ Š Œ}Œ}ŽîŽî�_‘_“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1‰}!1�‰}!1'¾!1 '¾!1ÅÂ!1ÅÂ!1[ç!1€[ç!1ù'"1üù'"1—,"1—,"1-Q"1p-Q"1Ë‘"1ìË‘"1i–"1ôi–"1Ú"1`Ú"1ßþ"1Ìßþ"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Ï› b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1_€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Ï› b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1_€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Ï› b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1_ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Hh  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss é é^^ÓÓ)H*H4»4»>.?.I¡I¡ST^‡^‡húhúrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðsséé__ÕÕ K K Á Á77­­##™™ƒƒøøm m"â"â$W%W'Ì'Ì)A*A,¶,¶.+/+1 1 346Š6Š8ÿ8ÿ:t;t=ç=ç?Z@ZBÍBÍD@E@G³G³I&J&L™L™N O QQSòSòUeVeXØXØZK[K]¾]¾_1`1b¤b¤degŠgŠiýiýkplpnãnãpVqVsÉsÉu^  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq é éaaÙÙ)Q*Q4É4É>???IµIµS+T+^¡^¡his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqééaaÙÙ Q Q É ÉAA¹¹11©©!!™™ ‰ ‰"#%y%y'ñ'ñ)i*i,á,á.Y/Y1Ñ1Ñ3I4I6Á6Á8999;±;±=)>)@¡@¡BCE‘E‘G H JJLõLõNkOkQáQáSWTWVÍVÍXCYC[¹[¹]/^/`¥`¥bce‘e‘ghj}j}lólónioiqßqßsUtUvËvËxAyA{·{·}-~-€£€£‚ƒ…�…�‡ˆŠ{Š{ŒñŒñŽg�g‘ݑݓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1ƒ}!1�ƒ}!1!¾!1 !¾!1ÅÂ!1ÅÂ!1Uç!1€Uç!1ó'"1üó'"1—,"1—,"1'Q"1p'Q"1Å‘"1ìÅ‘"1i–"1ôi–"1Ú"1`Ú"1ßþ"1Ìßþ"1#0#1ì#0#1ï~#1L ï~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%13€&1T 3€&1R'1˜R'1«d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ë— b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1^€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ë— b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1^€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ë— b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1^ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ii  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss é é\\ÏÏ)B*B4³4³>$?$I•I•ST^w^whèhèrWsW} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðsséé__ÕÕ K K Á Á77­­##™™  òòe e"Ø"Ø$K%K'¾'¾)1*1,¤,¤./1Š1Š3ý3ý5p6p8ã8ã:V;V=Ç=Ç?8@8B©B©DEG‹G‹IüIüKmLmNÞNÞPOQOSÀSÀU1V1X¢X¢Z[]„]„_õ_õafbfd×d×fHgHi¹i¹k*l*n›n›p q s}s}uîuîw_x_zÐzÐ|A}A²²�#‚#„”„”†‡‰s‰s‹ã‹ã�SŽS�Ã�Ã’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Ý}!1�Ý}!1{¾!1 {¾!1SÅ!1SÅ!1©ç!1€©ç!1G("1üG("1%/"1%/"1{Q"1p{Q"1’"1ì’"1÷˜"1ª÷˜"1…Ú"1š…Ú"1åþ"1Šåþ"1#0#1â#0#1ï~#1ºï~#1Ï$1’ Ï$1§ò$12 §ò$1áÞ%1R áÞ%1Õ€&1*Õ€&1¥R'1Ê¥R'1Me'1ê€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚"î b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1©é€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚"î b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1©é€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚"î bšoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r72w5:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1©é €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Oo  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ç ç]]ÓÓ)I*I4¿4¿>3?3I§I§ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqçç]]ÓÓ I I ¿ ¿55««!!——  ƒƒùùo o"å"å$[%['Ñ'Ñ)G*G,½,½.3/31©1©346•6•8 9 ;�;�=÷=÷?m@mBãBãDYEYGÏGÏICJCL·L·N+O+QŸQŸSTV‡V‡XûXûZo[o]ã]ã_W`WbËbËd?e?g³g³i'j'l›l›noqƒqƒs÷s÷ukvkxßxßzS{S}Ç}Ç;€;‚¯‚¯„#…#‡—‡—‰ Š ŒŒŽóŽó�g‘g“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1ƒ}!1�ƒ}!1!¾!1 !¾!1­Â!1­Â!1Oç!1€Oç!1í'"1üí'"1,"1,"1!Q"1p!Q"1¿‘"1ì¿‘"1Q–"1ôQ–"1sÚ"1`sÚ"1Óþ"1ÌÓþ"1#0#1ì#0#1ï~#1L ï~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Õ€&1T Õ€&1¥R'1˜¥R'1Me'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}̘ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}̘ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}̘ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Aa  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ê ê__ÔÔ)I*I4¾4¾>2?2I¦I¦ST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrêêbbÚÚ R R Ê ÊBBºº22ªª””  ~ ~"ó"ó$h%h'Ý'Ý)R*R,Ç,Ç.e>g²g²i&j&lšlšnoq‚q‚sösöujvjxÞxÞzR{R}Æ}Æ:€:‚®‚®„"…"‡–‡–‰ Š Œ€Œ€ŽõŽõ�j‘j“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1ã}!1�ã}!1�¾!1 �¾!1wÅ!1wÅ!1¯ç!1€¯ç!1M("1üM("1I/"1I/"1�Q"1p�Q"1’"1ì’"1™"1ô™"1mÚ"1`mÚ"1Íþ"1ÌÍþ"1#0#1ì#0#1ï~#1L ï~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1 �&1T  �&1ÛR'1˜ÛR'1ƒe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ+÷ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ+÷ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ+÷ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Õ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations19Y  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ä ä[[ÒÒ)I*I4À4À>5?5IªIªST^”^”h i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmää[[ÒÒ I I À À77®®%%œœŠŠ x x"ï"ï$f%f'Ý'Ý)T*T,Ë,Ë.B/B1¹1¹30406§6§89;•;•= > @ƒ@ƒBúBúDqEqGèGèI]J]LÒLÒNGOGQ¼Q¼S1T1V¦V¦XY[�[�]^`z`zbïbïddedgÙgÙiNjNlÃlÃn8o8q­q­s"t"v—v—x y {�{�}ö}ök€k‚à‚à„U…U‡ʇʉ?Š?Œ´Œ´Ž)�)‘ž‘ž“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1}}!1�}}!1¾!1 ¾!1­Â!1­Â!1Oç!1€Oç!1í'"1üí'"1,"1,"1!Q"1p!Q"1¿‘"1ì¿‘"1Q–"1ôQ–"1…Ú"1`…Ú"1åþ"1Ìåþ"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{È” b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{È” b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{È” b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Rr  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æ[[ÐÐ)E*E4º4º>-?-I I ST^„^„höhörhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  €€õõj j"ß"ß$T%T'É'É)>*>,³,³.(/(1�1�346‡6‡8ü8ü:q;q=æ=æ?[@[BÐBÐDEEEGºGºI-J-L L NOQ†Q†SùSùUlVlXßXßZR[R]Å]Å_8`8bªbªdegŽgŽijlrlrnänäpVqVsÈsÈu:v:x¬x¬z{}�}�€‚t‚t„æ„æ†X‡X‰ˉˋ>Œ>ޱޱ�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1ã}!1�ã}!1�¾!1 �¾!1wÅ!1wÅ!1¯ç!1€¯ç!1M("1üM("1I/"1I/"1�Q"1p�Q"1’"1ì’"1™"1ô™"1Ú"1`Ú"1ßþ"1Ìßþ"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹+÷ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ô€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹+÷ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ô€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹+÷ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ô €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Nn  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn ä äZZÏÏ)D*D4¹4¹>,?,IŸIŸST^…^…høhørksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnääZZÐÐ F F ¼ ¼22¨¨””  €€ööl l"â"â$W%W'Ì'Ì)A*A,¶,¶.+/+1 1 346Š6Š8ÿ8ÿ:t;t=é=é?^@^BÓBÓDHEHG½G½I0J0L£L£NOQ‰Q‰SüSüUoVoXâXâZU[U]È]È_;`;b®b®d!e!g”g”ijlzlzníníp`q`sÓsÓuFvFx¹x¹z,{,}Ÿ}Ÿ€‚…‚…„ø„ø†k‡k‰Þ‰Þ‹QŒQŽÄŽÄ�7‘7“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1}}!1�}}!1¾!1 ¾!1×Â!1×Â!1Iç!1€Iç!1ç'"1üç'"1©,"1©,"1Q"1pQ"1¹‘"1칑"1{–"1ô{–"1mÚ"1`mÚ"1Íþ"1ÌÍþ"10#1ì0#1é~#1L é~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1ù€&1T ù€&1ÉR'1˜ÉR'1qe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÉ• b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations16v€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÉ• b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations16v€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÉ• b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations16v €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1<\  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn å å\\ÓÓ)J*J4Á4Á>8?8I¯I¯S&T&^˜^˜h i s|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnåå\\ÓÓ J J Á Á88¯¯&&��‹‹ y y"ð"ð$g%g'Þ'Þ)U*U,Ì,Ì.C/C1º1º31416¨6¨89;–;–= > @„@„BûBûDrErGéGéI`J`L×L×NNONQÅQÅS1?1I¥I¥ST^�^�hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmççaaÛÛ U U Ï ÏIIÃÃ==··--££ � �"#%{%{'ñ'ñ)g*g,Ý,Ý.S/S1É1É3?4?6µ6µ8+9+;¡;¡=>@‰@‰BýBýDqEqGåGåIYJYLÍLÍNAOAQµQµS)T)V�V�XY[…[…]ù]ù_m`mbábádUeUgÉgÉi=j=l±l±n%o%q™q™s t v�v�xõxõzi{i}Ý}ÝQ€Q‚łń9…9‡«‡«‰ŠŒ‘Œ‘Ž�‘w‘w“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1w}!1�w}!1¾!1 ¾!1×Â!1×Â!1Iç!1€Iç!1ç'"1üç'"1©,"1©,"1Q"1pQ"1¹‘"1칑"1{–"1ô{–"1Ú"1`Ú"1ßþ"1Ìßþ"10#1ì0#1é~#1L é~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÅ‘ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations17w€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÅ‘ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations17w€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÅ‘ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations17w €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Qq  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll á áVVËË)@*@4µ4µ>*?*IŸIŸST^†^†høhørjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllááVVËË @ @ µ µ**ŸŸ‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?N@NBÃBÃD8E8G­G­I"J"L—L—N O Q�Q�SöSöUkVkXàXàZU[U]Ê]Ê_?`?b´b´d)e)gžgžijlˆlˆnúnúplqlsÞsÞuPvPxÂxÂz4{4}¦}¦€‚ЂЄü„ü†n‡n‰à‰à‹RŒRŽÄŽÄ�6‘6“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ã}!1�ã}!1�¾!1 �¾!1;Å!1;Å!1µç!1€µç!1S("1üS("1 /"1 /"1‡Q"1p‡Q"1%’"1ì%’"1ߘ"1ôߘ"1ÕØ"1`ÕØ"15ý"1Ì5ý"10#1ì0#1é~#1L é~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1i€&1T i€&19R'1˜9R'1ád'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”4 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1™Ù€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”4 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1™Ù€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”4 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1™Ù €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Ç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm å åYYÍÍ)A*A4³4³>%?%I—I—S T ^{^{híhír]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmåå]]ÕÕ M M Å Å==µµ--¥¥�� u u"é"é$]%]'Ñ'Ñ)E*E,¹,¹.-/-1¡1¡346‰6‰8ý8ý:q;q=ã=ã?U@UBÇBÇD9E9G«G«IJL�L�NOQsQsSåSåUWVWXÉXÉZ;[;]­]­_`b‘b‘deguguiçiçkYlYnËnËp=q=s¯s¯u!v!x“x“z{}w}wéé�[‚[„̈́͆=‡=‰®‰®‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1q}!1�q}!1¾!1 ¾!1­Â!1­Â!1Cç!1€Cç!1á'"1üá'"1,"1,"1Q"1pQ"1³‘"1쳑"1Q–"1ôQ–"1aÚ"1`aÚ"1Áþ"1ÌÁþ"10#1ì0#1ã~#1L ã~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwÆ’ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwÆ’ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwÆ’ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations15U  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn è èbbØØ)N*N4Ä4Ä>8?8I¬I¬S T ^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnèèbbÜÜ V V Ð ÐJJÄÄ>>¸¸22¬¬&&    "#%�%�'(*|*|,ò,ò.h/h1Þ1Þ3T4T6Ê6Ê8@9@;¶;¶=,>,@¢@¢BCEŽEŽGHJxJxLìLìN`O`QÔQÔSHTHV¼V¼X0Y0[¤[¤]^`Œ`Œbcetetgègèi\j\lÐlÐnDoDq¸q¸s,t,v v xy{ˆ{ˆ}ü}üp€p‚ä‚ä„X…X‡̇̉@Š@Œ´Œ´Ž(�(‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1é}!1�é}!1‡¾!1 ‡¾!1qÅ!1qÅ!1»ç!1€»ç!1Y("1üY("1C/"1C/"1�Q"1p�Q"1+’"1ì+’"1™"1ô™"1ÕØ"1`ÕØ"15ý"1Ì5ý"10#1ì0#1ã~#1L ã~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1)�&1T )�&1ùR'1˜ùR'1¡e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jž=  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ò€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jž=  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ò€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jž=  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ò €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Ç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii â â[[ÔÔ)M*M4Æ4Æ>=?=I´I´S)T)^œ^œhis€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiââ[[ÔÔ M M Æ Æ??¸¸11ªª##œœ Ž Ž"#%€%€'ù'ù)r*r,ë,ë.d/d1Ý1Ý3V4V6Ï6Ï8H9H;Á;Á=:>:@³@³B,C,E¥E¥GHJ•J•L M OƒOƒQúQúSqTqVèVèX_Y_[Ö[Ö]M^M`Ä`Äb9c9e®e®g#h#j˜j˜l m o€o€qóqósftfvÙvÙxLyL{¿{¿}2~2€¥€¥‚ƒ…‹…‹‡ü‡ü‰nŠnŒàŒàŽR�R‘đē b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1q}!1�q}!1¾!1 ¾!1­Â!1­Â!1=ç!1€=ç!1Û'"1üÛ'"1,"1,"1Q"1pQ"1­‘"1ì­‘"1Q–"1ôQ–"1Ú"1`Ú"1ßþ"1Ìßþ"10#1ì0#1ã~#1L ã~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÂŽ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÂŽ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÂŽ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Rr  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ä ä\\ÐÐ)D*D4¸4¸>*?*IœIœST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllää\\ÔÔ L L Ä Ä<<´´,,¤¤””  „ „"ü"ü$p%p'ä'ä)X*X,Ì,Ì.@/@1´1´3(4(6œ6œ89;„;„=ø=ø?l@lBàBàDTETGÈGÈI:J:L¬L¬NOQ�Q�STVtVtXæXæZX[X]Ê]Ê_<`v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1é}!1�é}!1‡¾!1 ‡¾!1kÅ!1kÅ!1»ç!1€»ç!1Y("1üY("1=/"1=/"1�Q"1p�Q"1+’"1ì+’"1™"1ô™"1ÕØ"1`ÕØ"15ý"1Ì5ý"10#1ì0#1ã~#1L ã~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1·€&1T ·€&1‡R'1˜‡R'1/e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Ç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii à àWWÎÎ)E*E4¼4¼>1?1I¦I¦ST^Š^Šhûhûrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiààWWÎÎ E E ¼ ¼33ªª!!˜˜††ýýt t"ë"ë$b%b'Ù'Ù)P*P,Ç,Ç.>/>1µ1µ3,4,6£6£89;‘;‘=>@@BöBöDmEmGäGäIYJYLÎLÎNCOCQ¸Q¸S-T-V¢V¢XY[Œ[Œ]^`v`vbébéd\e\gÏgÏiBjBlµlµn&o&q—q—stvyvyxêxêz[{[}Ì}Ì=€=‚®‚®„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1k}!1�k}!1 ¾!1  ¾!1ËÂ!1ËÂ!1=ç!1€=ç!1Û'"1üÛ'"1�,"1�,"1Q"1pQ"1­‘"1ì­‘"1o–"1ôo–"1ÏØ"1`ÏØ"1/ý"1Ì/ý"10#1ì0#1ã~#1L ã~#1¥$1¬ ¥$1}ò$1p }ò$1·Þ%1t ·Þ%1{€&1T {€&1KR'1˜KR'1ód'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÃ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations13s€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÃ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations13s€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÃ� b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations13s €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Ç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝRRÇÇ)<*<4¯4¯>"?"I•I•ST^{^{hîhîr_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��wwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Û=Û?N@NBÁBÁD4E4G§G§IJL�L�NOQsQsSæSæUYVYXÌXÌZ?[?]²]²_%`%b˜b˜d e g~g~iñiñkdldn×n×pJqJs½s½u0v0x£x£z{}‰}‰üü�o‚o„â„â†S‡S‰ʼnŋ7Œ7ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1ï}!1�ï}!1�¾!1 �¾!1/Å!1/Å!1Áç!1€Áç!1_("1ü_("1/"1/"1“Q"1p“Q"11’"1ì1’"1Ó˜"1ôÓ˜"1ÏØ"1`ÏØ"1/ý"1Ì/ý"10#1ì0#1ã~#1L ã~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1/�&1T /�&1ÿR'1˜ÿR'1§e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§F b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1™Ù€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§F b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1™Ù€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§F b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1™Ù €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Ç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝUUÍÍ)E*E4½4½>3?3I©I©ST^•^•h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%���� } }"õ"õ$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=>@•@•B C E…E…GýGýIsJsLéLéN_O_QÕQÕSKTKVÁVÁX7Y7[­[­]#^#`™`™bce…e…gûgûiqjqlçlçn]o]qÓqÓsItIv¿v¿x5y5{«{«}!~!€—€—‚ ƒ …ƒ…ƒ‡ù‡ù‰oŠoŒåŒåŽ[�[‘ёѓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1k}!1�k}!1 ¾!1  ¾!1ËÂ!1ËÂ!17ç!1€7ç!1Õ'"1üÕ'"1�,"1�,"1 Q"1p Q"1§‘"1ì§‘"1o–"1ôo–"1ÏØ"1`ÏØ"1/ý"1Ì/ý"10#1ì0#1ã~#1L ã~#1“$1¬ “$1kò$1p kò$1¥Þ%1t ¥Þ%1€&1T €&1åQ'1˜åQ'1�d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r¿‹ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations14t€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r¿‹ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations14t€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>r¿‹ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations14t €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Ç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝPPÃÃ)6*64§4§>?I‰I‰SúSú]k^khÜhÜrKsK} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��ssææY Y"Ì"Ì$?%?'²'²)%*%,˜,˜. / 1~1~3ñ3ñ5d6d8×8×:J;J=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éaZbZdËdËf'?'I›I›ST^ƒ^ƒh÷h÷rksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡‹‡‹‰ÿ‰ÿ‹sŒsŽçŽç�[‘[“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1e}!1�e}!1¾!1 ¾!1¡Â!1¡Â!17ç!1€7ç!1Õ'"1üÕ'"1s,"1s,"1 Q"1p Q"1§‘"1ì§‘"1E–"1ôE–"1ÉØ"1`ÉØ"1)ý"1Ì)ý"10#1ì0#1Ý~#1L Ý~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1½€&1T ½€&1�R'1˜�R'15e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÀŒ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÀŒ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÀŒ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Ç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞSSÈÈ)=*=4²4²>&?&IšIšST^„^„høhørlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžˆˆýýr r"ç"ç$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`JbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡Œ‡Œ‰ŠŒvŒvŽëŽë�`‘`“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1õ}!1�õ}!1“¾!1 “¾!1eÅ!1eÅ!1Çç!1€Çç!1e("1üe("17/"17/"1™Q"1p™Q"17’"1ì7’"1 ™"1ô ™"1Ú"1`Ú"1ßþ"1Ôßþ"10#1ô0#1Ý~#1T Ý~#1Ã$1´ Ã$1ëð$1x ëð$1ÉÀ%1| ÉÀ%1ù€&1\ ù€&1ïP'1 ïP'1—c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°O b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ï€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°O b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ï€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°O b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ï €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1¬„Uy  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>)?)IžIžST^’^’his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜIQJQLÆLÆN;O;Q°Q°S%T%VšVšXY[„[„]ù]ù_n`nbãbãdXeXgÍgÍiBjBl·l·n6o6q«q«s t v•v•x y {{}ô}ôi€i‚Þ‚Þ„S…S‡ȇȉ=Š=Œ²Œ²Ž'�'‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1_}!1�_}!1ý½!1 ý½!1¡Â!1¡Â!11ç!1€1ç!1Ï'"1üÏ'"1s,"1s,"1Q"1pQ"1¡‘"1ì¡‘"1E–"1ôE–"1ÉØ"1`ÉØ"1)ý"1Ì)ý"10#1ì0#1Ý~#1L Ý~#1¥$1¬ ¥$1}ò$1p }ò$1·Þ%1t ·Þ%1u€&1T u€&1ER'1˜ER'1íd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¼ˆ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¼ˆ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¼ˆ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1§Ç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94®4®>!?!I”I”S T ^|^|hîhîr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?O@OBÄBÄD9E9G®G®I!J!L”L”NOQzQzSíSíU`V`XÓXÓZF[F]¹]¹_,`,b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†P‡P‰ÉË6Œ6ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1õ}!1�õ}!1“¾!1 “¾!1eÅ!1eÅ!1Çç!1€Çç!1e("1üe("17/"17/"1™Q"1p™Q"17’"1ì7’"1 ™"1ô ™"1ÇÚ"1`ÇÚ"1'ÿ"1Ô'ÿ"10#1ô0#1Ý~#1T Ý~#1«$1´ «$1Óð$1x Óð$1±À%1| ±À%1�€&1\ �€&1ƒP'1 ƒP'1+c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯O b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1΀} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯O b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1΀} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯O b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Î €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1É¡¡Å  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÃÃ)8*84­4­> ? I“I“ST^�^�hôhôrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD.?.I¥I¥ST^š^šh i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLÍLÍNDODQ»Q»S2T2V©V©X Y [—[—]^`…`…bübüdsesgêgêiajalØlØnVoVqÈqÈs:t:v¬v¬xy{�{�}~€t€t‚æ‚æ„X…X‡ʇʉ<Š<Œ®Œ®Ž � ‘’‘’“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L%i!1û}!1�û}!1™¾!1 ™¾!1)Å!1)Å!1Çç!1€Çç!1e("1üe("1û."1û."1™Q"1p™Q"17’"1ì7’"1͘"1ô͘"1¥Ø"1`¥Ø"1ý"1Ìý"1 0#1ì 0#1×~#1L ×~#1½$1¬ ½$1åð$1p åð$1ÃÀ%1t ÃÀ%1Û€&1T Û€&1ÑP'1˜ÑP'1yc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹X$ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1™Ù€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹X$ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1™Ù€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹X$ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1™Ù €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1§£  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Û ÛQQÑÑ)G*G4»4»>/?/I£I£S#T#^—^—h i s}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÛÛUUÏÏ I I à Ã==··11««!!——  ƒ ƒ"ù"ù$y%y'ï'ï)e*e,Û,Û.Q/Q1Ç1Ç3=4=6³6³8)9);Ÿ;Ÿ=>@‡@‡BûBûDoEoGãGãIWJWLËLËN?O?Q³Q³S'T'V›V›XY[ƒ[ƒ]÷]÷_k`kbëbëd_e_gÓgÓiGjGl»l»n/o/q£q£stv‹v‹xÿxÿzs{s}ç}ç[€[‚ςτC…C‡µ‡µ‰(Š(Œ›Œ›Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L%i!1Y}!1�Y}!1÷½!1 ÷½!1¿Â!1¿Â!1+ç!1€+ç!1É'"1üÉ'"1‘,"1‘,"1ýP"1pýP"1›‘"1웑"1c–"1ôc–"1‹Ú"1`‹Ú"1ëþ"1Ôëþ"1 0#1ô 0#1×~#1T ×~#1±$1´ ±$1Ùð$1x Ùð$1·À%1| ·À%1“€&1\ “€&1‰P'1 ‰P'11c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l¹… b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations11q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l¹… b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations11q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations18l¹… b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations11q €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1´Œg‹  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Õ ÕJJÃÃ)8*84­4­>"?"I—I—S T ^Š^Šhühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÕÕJJ¿¿ 4 4 © ©““}}òòggÜÜQ Q"Æ"Æ$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nþnþppqpsâsâuTvTxÆxÆz8{8}ª}ª€‚ނބ…‡r‡r‰ä‰ä‹VŒVŽÈŽÈ�:‘:“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L%i!1û}!1�û}!1™¾!1 ™¾!1)Å!1)Å!1Çç!1€Çç!1e("1üe("1û."1û."1™Q"1p™Q"17’"1ì7’"1͘"1ô͘"1¥Ø"1`¥Ø"1ý"1Ìý"1 0#1ì 0#1×~#1L ×~#1¥$1¬ ¥$1Íð$1p Íð$1«À%1t «À%1c€&1T c€&1YP'1˜YP'1c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸X$ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1™Ù€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸X$ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1™Ù€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸X$ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1™Ù €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1§£  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ù ÙMMÉÉ)=*=4¯4¯>!?!I“I“ST^ƒ^ƒhõhõrese} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A ¹ ¹11©©!!™™  ��õõi i"Ý"Ý$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b™b™d e g}g}iïiïkalanÓnÓpEqEs·s·u)v)x›x›z { }}ññ�c‚c„Õ„Õ†E‡E‰¶‰¶‹'Œ'Ž˜Ž˜� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„i!15}!15}!1Ó½!1|Ó½!17Ã!1„7Ã!1mÆ!1ðmÆ!1÷Ï!1|÷Ï!1-Ó!1H-Ó!1ç!1Ôç!1¥'"1P¥'"1 -"1X -"1?0"1Ä?0"1É9"1PÉ9"1ÿ<"1ÿ<"1ÙP"1¨ÙP"1w‘"1$ w‘"1Û–"1, Û–"1š"1˜ š"1›£"1$ ›£"1Ѧ"1ð Ѧ"1Ô"1| Ô"1‹Ú"1X‹Ú"1ëþ"1Ôëþ"1-,#1ô-,#1õ9#1Ðõ9#1±n#1\±n#1QŒ#1€QŒ#1O—#1¬O—#1Ï$18Ï$1§ò$1ü§ò$1áÞ%1áÞ%1[ì%1à[ì%15�&1l5�&1S'1°S'1­e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¡m b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1©é bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¡m b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1©é  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¡m  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1©é bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¬Ì©Év– bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1)i9y²ò+k¤ä  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d)©9¹òrîn«+³3däxø�ÖV�HÈ�º:tô-­æf Ÿ  X Ø  ‘ Ê C ƒ z º ¼ ü 5 u  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10CC × ×OOÑÑ)I*I4Á4Á>9?9I¯I¯S%T%^§^§his“s“} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðCC»»33«« # # · ·//§§——‡‡ÿÿw w"ï"ï$q%q'é'é)a*a,Ù,Ù.Q/Q1É1É3A4A6¹6¹8191;©;©=!>!@™@™BCE‰E‰GHJyJyLïLïNeOeQÛQÛSQTQVÇVÇX=Y=[³[³])^)`Ÿ`Ÿbce‹e‹ghjwjwlílínoooqåqås[t[vÑvÑxGyG{½{½}3~3€©€©‚ƒ…•…•‡ ˆ Š�Š�Œ÷Œ÷Žm�m‘ã‘ã“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxi!1~!1è~!1¥¾!1d¥¾!1MÅ!1lMÅ!1×Î!1è×Î!1 Ò!1´ Ò!1Ùç!1@Ùç!1w("1¼w("1/"1Ä/"1©8"1@©8"1ß;"1 ß;"1«Q"1˜«Q"1I’"1I’"1ñ˜"1 ñ˜"1{¢"1˜ {¢"1±¥"1d ±¥"1óÒ"1ð óÒ"1gÚ"1Ì gÚ"1Çþ"1H Çþ"1 ,#1h ,#11:#1D1:#1ín#1Ðín#1}�#1ô}�#1{˜#1 {˜#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1ç€&1àç€&1·R'1$·R'1_e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ñn: b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 E…­í¾þ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ñn: b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 E…­í¾þ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ñn:  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 E…­í¾þ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ž®Mm?_ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations11qG‡ÎW—à  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d1±GÇŽü|ÕUÇGœ’cã*ªñq¸8�JÊ“Ü\ ¥ % n î 7 ·  € Ç  á ! N Ž × `  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÉÉ)@*@4›4›>0?0I§I§ST^œ^œhis€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=ç=ç?^@^BÕBÕDLELGÃGÃIXJXLÏLÏNFOFQ½Q½S4T4V«V«X"Y"[™[™]^`‡`‡bþbþdueugìgìicjclÚlÚnXoXqÊqÊsŠ>Œ°Œ°Ž"�"‘”‘”“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„i!15}!15}!1Ó½!1|Ó½!17Ã!1„7Ã!1mÆ!1ðmÆ!1÷Ï!1|÷Ï!1-Ó!1H-Ó!1ç!1Ôç!1Ÿ'"1PŸ'"1 -"1X -"1?0"1Ä?0"1É9"1PÉ9"1ÿ<"1ÿ<"1ÓP"1¨ÓP"1q‘"1$ q‘"1Û–"1, Û–"1š"1˜ š"1›£"1$ ›£"1Ѧ"1ð Ѧ"1Ô"1| Ô"1UÚ"1XUÚ"1µþ"1Ôµþ"1÷+#1ô÷+#1:#1Ð:#1;o#1\;o#1�#1€�#1 š#1¬ š#1·$18·$1�ò$1ü�ò$1ÉÞ%1ÉÞ%1Cì%1àCì%1½€&1l½€&1�R'1°�R'15e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ožj b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1©é bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ožj b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1©é  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Ožj  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1©é bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 šº©ÉDd bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1]�®îú:•Õ0p  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d]Ý:º—ôt=½ÑQ‚®.ÇG‹ Œ hèCÞùy Ô T ¯ / Š e å ÿ ? @ À ]ºúU•ð0 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10BB Õ ÕKKÉÉ)?*?4µ4µ>+?+IŸIŸST^“^“his{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðBB¸¸..¤¤   ­ ­##™™……ûûqqçç] ]"Ó"Ó$Q%Q'Ç'Ç)=*=,³,³.)/)1Ÿ1Ÿ346‹6‹89;w;w=í=í?c@cBÙBÙDOEOGÅGÅI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÃbÃd7e7g«g«ijl“l“noq‡q‡sûsûuovoxãxãzW{W}Ë}Ë?€?‚³‚³„'…'‡›‡›‰ŠŒƒŒƒŽ÷Ž÷�k‘k“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxi!1 ~!1è ~!1«¾!1d«¾!1MÅ!1lMÅ!1×Î!1è×Î!1 Ò!1´ Ò!1Ùç!1@Ùç!1w("1¼w("1/"1Ä/"1©8"1@©8"1ß;"1 ß;"1«Q"1˜«Q"1I’"1I’"1ñ˜"1 ñ˜"1{¢"1˜ {¢"1±¥"1d ±¥"1óÒ"1ð óÒ"1…Ú"1Ì …Ú"1åþ"1H åþ"1',#1h',#1U:#1DU:#1o#1Ðo#11Ž#1ô1Ž#1/™#1 /™#1«$1¬«$1ƒò$1pƒò$1½Þ%1t½Þ%17ì%1T7ì%1�€&1à�€&1]R'1$]R'1e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒp< b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 F†®î¿ÿ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒp< b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 F†®î¿ÿ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒp<  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 F†®î¿ÿ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¡Á>r’ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1R’¼üä$t´D  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥z©*©b­reasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dRÒ$¤övÈH]Ýš(¨lìós>¾¾>�à`°0€ P Ð    ð p À @ �  b ¢ l ¬ ô 4„ÄT bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Õ ÕJJÃÃ)8*84�4�>$?$I™I™ST^Œ^Œhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÕÕJJ¿¿ 4 4 © ©““}}òòggÜÜQ Q"Æ"Æ$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=´=´?)@)BžBžDEGˆGˆIJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚noqrqrsäsäuVvVxÈxÈz:{:}¬}¬€‚�‚�„…‡t‡t‰æ‰æ‹XŒXŽÊŽÊ�<‘<“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1_}!1è_}!1ý½!1dý½!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!1+ç!1@+ç!1É'"1¼É'"1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1ýP"1˜ýP"1›‘"1›‘"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"1‘Ú"1Ì ‘Ú"1ñþ"1H ñþ"13,#1h3,#1û9#1Dû9#1·n#1зn#1]Œ#1ô]Œ#1[—#1 [—#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1ó€&1àó€&1ÃR'1$ÃR'1ke'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¾Š b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 <|9yÈ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¾Š b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 <|9yÈ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¾Š  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 <|9yÈ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ©É¯Ïx˜ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations19y"b²ò+k¤ä  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d9¹"¢òréi«+°0däw÷�ÖV‘LÌ…¾>w÷0°éi ¢ " [ Û  – Ï  H ˆ ” Ô Á  : z  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Ö ÖMMÄÄ);*;4²4²> ? IžIžST^’^’his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÖÖMMÄÄ ; ; ² ²))  ŽŽ||óój j"á"á$X%X'Ï'Ï)F*F,½,½.4/41«1«3"4"6™6™89;‡;‡=þ=þ?u@uBìBìDcEcGÚGÚI5J5LªLªNOQ”Q”S T VšVšXY[„[„]ù]ù_n`nbãbãdXeXgÍgÍiBjBl·l·n6o6q«q«s t v•v•x y {{}ô}ôi€i‚Þ‚Þ„S…S‡ȇȉ=Š=Œ²Œ²Ž'�'‘œ‘œ“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1Ÿ¾!1dŸ¾!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!1Óç!1@Óç!1q("1¼q("1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1¥Q"1˜¥Q"1C’"1C’"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"1Ú"1Ì Ú"1ßþ"1H ßþ"1!,#1h!,#1+4#1D+4#1Õw#1ÀÕw#1Ù‹#1äÙ‹#1×–#1À×–#1“$1L“$1kò$1kò$1¥Þ%1¥Þ%1ì%1ôì%1É€&1€É€&1™R'1Ä™R'1Ae'1È€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Æc/ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 P�—×É  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Æc/ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 P�—×É  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Æc/  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 P�—×É   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �¯Aab‚ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 Hˆ,lÐ bÒ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1<HÈ,¬�õuØX¾>  ‡hè0°øxÀ@ˆÈP¼ ü  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff à àZZÐÐ)F*F4¼4¼>2?2I¦I¦ST^�^�his\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffààZZÔÔ N N È ÈBB¼¼66°°**¤¤ ˜ ˜"#%ˆ%ˆ'þ'þ)t*t,ê,ê.`/`1Ö1Ö3L4L6Â6Â8898;®;®=$>$@š@šBCE†E†GüGüIrJrLæLæNZOZQÎQÎSBTBV¶V¶X*Y*[ž[ž]^`†`†búbúdnengâgâiVjVlÊlÊn@o@q´q´s(t(vœvœxy{„{„}ø}øl€l‚à‚à„T…T‡¬‡¬‰ Š Œ”Œ”Ž�‘|‘|“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1Y}!1èY}!1÷½!1d÷½!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!1+ç!1@+ç!1É'"1¼É'"1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1ýP"1˜ýP"1›‘"1›‘"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"1[Ú"1Ì [Ú"1»þ"1H »þ"1ý+#1hý+#1û9#1Dû9#1·n#1зn#1WŒ#1ôWŒ#1U—#1 U—#1¥$1¬¥$1}ò$1p}ò$1·Þ%1t·Þ%11ì%1T1ì%1‡€&1à‡€&1WR'1$WR'1ÿd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¼ˆ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ;{9yÇ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¼ˆ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ;{9yÇ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¼ˆ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ;{9yÇ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ©É™¹Bb bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations19y²ò©é+k¤ä  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d9¹òr«+dä�>¾×W…�ÌLIÉ“‚»;tô-­æf Ÿ  X Ø  ‘ Ê C ƒ ¼ ü  [ 5 u  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ö ÖLLÁÁ)6*64«4«>?I“I“ST^�^�hôhôrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††üürrèè^ ^"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯IJLzLzNíNíP`Q`SÓSÓUcVcXÖXÖZI[I]¼]¼_/`/b¢b¢degˆgˆiûiûknlnnénép\q\sÏsÏuBvBxµxµz({(}›}›€‚�‚�„ô„ô†g‡g‰Ú‰Ú‹MŒMŽÀŽÀ�3‘3“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1Ÿ¾!1dŸ¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1Óç!1@Óç!1q("1¼q("11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1¥Q"1˜¥Q"1C’"1C’"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1‹Ú"1Ì ‹Ú"1ëþ"1H ëþ"1-,#1h-,#15#1D5#1­x#1À­x#1cŒ#1äcŒ#1a—#1Àa—#1{$1L{$1Sò$1Sò$1�Þ%1�Þ%1ì%1ôì%1W€&1€W€&1'R'1Ä'R'1Ïd'1È€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Çe1 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘—×Ê  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Çe1 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘—×Ê  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Çe1  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘—×Ê   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¢ÂAat” bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 9y²öµõ bÒ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1<9¹òr«+dä�PÐØX™‘âbJÊ«+C|¼E …  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜTTÈÈ)<*<4°4°>$?$I–I–ST^~^~hðhðrEsE} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœŒŒ | |"ô"ô$h%h'Ü'Ü)P*P,Ä,Ä.8/81¬1¬3 4 6”6”89;|;|=ð=ð?d@dBØBØDLELGÀGÀI4J4L¦L¦NOQŠQŠSüSüUnVnXàXàZR[R]Ä]Ä_6`6b¨b¨degŒgŒiþiþkplpnænæpXqXsÊsÊu)?)I�I�ST^…^…hùhùrQsQ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÝÝWWÑÑ K K Å Å??¹¹33­­##™™ … …"û"û$q%q'ç'ç)]*],Ó,Ó.I/I1¿1¿35456«6«8!9!;—;—= > @�@�BõBõDiEiGÝGÝIQJQLÅLÅN9O9Q­Q­S!T!V•V•X Y [}[}]ñ]ñ_e`ebÙbÙdMeMgÁgÁi5j5l©l©noq‘q‘stvyvyxíxíza{a}Õ}ÕI€I‚½‚½„1…1‡‰‡‰‰ü‰ü‹oŒoŽâŽâ�U‘U“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l%i!1û}!1Ðû}!1™¾!1L™¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Íç!1(Íç!1k("1¤k("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1ŸQ"1€ŸQ"1=’"1ü=’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#1u2#1,u2#1¯‹#1˜¯‹#1Ÿ$1$Ÿ$1wò$1èwò$1±Þ%1ì±Þ%1u€&1Ìu€&1ER'1ER'1íd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»X$ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [›ØÐ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»X$ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [›ØÐ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»X$  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [›ØÐ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ©Éˆ¨3S bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1_€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ÿæf­-tô� Ï  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84­4­> ? I“I“ST^y^yhìhìr_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?L@LB¿B¿D2E2G¥G¥IJL‹L‹NþNþPqQqSäSäUWVWXÊXÊZ=[=]°]°_#`#b–b–d e g|g|iïiïkblbnÕnÕpHqHs»s»u.v.x¡x¡z{}‡}‡úú�m‚m„à„à†S‡S‰ʼnŋ7Œ7ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1_}!1è_}!1ý½!1dý½!1kÅ!1lkÅ!1õÎ!1èõÎ!1+Ò!1´+Ò!11ç!1@1ç!1Ï'"1¼Ï'"1=/"1Ä=/"1Ç8"1@Ç8"1ý;"1 ý;"1Q"1˜Q"1¡‘"1¡‘"1™"1 ™"1™¢"1˜ ™¢"1Ï¥"1d Ï¥"1Ó"1ð Ó"1UÚ"1Ì UÚ"1µþ"1H µþ"1÷+#1h÷+#1÷4#1D÷4#1¡x#1À¡x#1QŒ#1äQŒ#1O—#1ÀO—#1u$1Lu$1Mò$1Mò$1‡Þ%1‡Þ%1ì%1ôì%13€&1€3€&1R'1ÄR'1«d'1È€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n½‰ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U9yÒ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n½‰ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U9yÒ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n½‰  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U9yÒ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 –¶ <\ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 G‡Î¢â bÒ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1<GÇŽÕUœcã=½*ª†ñqÏO¸8˜¿F2 r  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛOOÃÃ)7*74«4«>?I�I�ST^s^shåhår:s:} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››ƒƒ÷÷k k"ß"ß$S%S'Ç'Ç);*;,¯,¯.#/#1—1—3 4 668ó8ó:g;g=Û=Û?M@MB¿B¿D1E1G£G£IJL‡L‡NùNùPkQkSÝSÝUOVOXÁXÁZ3[3]¥]¥_`b‰b‰dûdûfmgmißißkQlQnÃnÃp5q5s§s§uvx‹x‹zýzý|o}oáá�S‚S„ńņ‡‰‹‰‹‹ü‹ü�mŽm�Þ�Þ’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l%i!1û}!1Ðû}!1™¾!1L™¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Íç!1(Íç!1k("1¤k("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1ŸQ"1€ŸQ"1=’"1ü=’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1÷4#1,÷4#13Œ#1˜3Œ#1�$1$�$1eò$1èeò$1ŸÞ%1ìŸÞ%1€&1Ì€&1ßQ'1ßQ'1‡d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼Z& b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 \œÚÑ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼Z& b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 \œÚÑ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼Z&  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 \œÚÑ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ©É›»f† bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¦æ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1;»‚ÉI� V  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙLL¿¿)2*24¥4¥>?I‡I‡SøSø]i^ihÚhÚrKsK} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰üüooââU U"È"È$;%;'®'®)!*!,”,”./1z1z3í3í5`6`8Ó8Ó:F;F=¹=¹?*@*B›B›D E G}G}IîIîK_L_NÐNÐPAQAS²S²U#V#X”X”Z[]v]v_ç_çaXbXdÉdÉf:g:i«i«kln�n�pþpþrosouàuàwQxQzÂzÂ|3}3¤¤�‚„†„††÷†÷ˆg‰g‹׋×�GŽG�·�·’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1e}!1Ðe}!1¾!1L¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!17ç!1(7ç!1Õ'"1¤Õ'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1 Q"1€ Q"1§‘"1ü§‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1i2#1,i2#1å‹#1˜å‹#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1�&1Ì�&1íR'1íR'1•e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;oÀŒ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 1q;{à  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;oÀŒ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 1q;{à  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;oÀŒ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 1q;{à   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ÿ¿HhXx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1P€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�Û[¦&qñÀ  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ü ÜUUÎÎ)G*G4À4À>9?9I°I°S%T%^˜^˜h i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÜÜUUÎÎ G G À À99²²++¤¤–– ˆ ˆ"#%z%z'ó'ó)l*l,å,å.^/^1×1×3P4P6É6É8B9B;»;»=4>4@­@­B&C&EŸEŸGHJ‘J‘LMOOQöQöSmTmVäVäX[Y[[Ò[Ò]I^I`À`Àb5c5eªeªghj”j”l m o|o|qïqïsbtbvÕvÕxHyH{»{»}.~.€¡€¡‚ƒ…‡…‡‡ú‡ú‰lŠlŒތގP�P‘‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1õ}!1Ðõ}!1“¾!1L“¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Áç!1(Áç!1_("1¤_("1[/"1¬[/"1å8"1(å8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1CÚ"1´ CÚ"1£þ"10 £þ"1å+#1På+#1i2#1,i2#1…‹#1˜…‹#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°M b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 f¦—×î. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°M b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 f¦—×î. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°M  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 f¦—×î.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ©É}�!A bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1R€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1’×Wœaáb ¢  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞSSÈÈ)=*=4²4²>(?(IœIœST^„^„høhørnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžˆˆýýr r"ç"ç$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡އމŠŒxŒxŽíŽí�b‘b“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1e}!1Ðe}!1¾!1L¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!17ç!1(7ç!1Õ'"1¤Õ'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1 Q"1€ Q"1§‘"1ü§‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#1ñ4#1,ñ4#1oŒ#1˜oŒ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1«€&1Ì«€&1{R'1{R'1#e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¾Š b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 0p:zÞ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¾Š b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 0p:zÞ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;o¾Š  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 0p:zÞ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‰©%E"B bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1›Û€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1<¼‡ÒR�K ‹  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>-?-I¢I¢ST^†^†h÷h÷risi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbåbådXeXgËgËi>j>l±l±n"o"q“q“stvuvuxæxæzW{W}È}È9€9‚ª‚ª„…‡�‡�‰þ‰þ‹oŒoŽàŽà�Q‘Q“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1õ}!1Ðõ}!1“¾!1L“¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Çç!1(Çç!1e("1¤e("1[/"1¬[/"1å8"1(å8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1ñ4#1,ñ4#1 Œ#1˜ Œ#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1{€&1Ì{€&1KR'1KR'1ód'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±O b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 g§—×ð0 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±O b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 g§—×ð0 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±O  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 g§—×ð0  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ©É”´Tt bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1—×€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1&¦ëk°0uõç '  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94®4®>#?#I–I–ST^z^zhìhìr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?O@OBÄBÄD9E9G®G®I#J#L–L–N O Q|Q|SïSïUbVbXÕXÕZH[H]»]»_.`.b b deg„g„iöiökhlhnÚnÚpLqLs¾s¾u0v0x¢x¢z{}†}†øø�j‚j„܄܆P‡P‰ÉË6Œ6ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1k}!1Ðk}!1 ¾!1L ¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1=ç!1(=ç!1Û'"1¤Û'"1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1Q"1€Q"1­‘"1ü­‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1c2#1,c2#1ý‹#1˜ý‹#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1/�&1Ì/�&1ÿR'1ÿR'1§e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<pÁ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IQ‘@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<pÁ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IQ‘@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<pÁ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IQ‘@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 –¶QqBb bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1G€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1‡ÔT¡!nî×   €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSËË)C*C4»4»>3?3I©I©ST^•^•h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$k%k'ã'ã)[*[,Ó,Ó.K/K1Ã1Ã3;4;6³6³8+9+;£;£=>@“@“B C EƒEƒGûGûIsJsLéLéN_O_QÕQÕSKTKVÁVÁX7Y7[­[­]#^#`™`™bce…e…gûgûiqjqlçlçn]o]qÓqÓsItIv¿v¿x5y5{«{«}!~!€—€—‚ ƒ …ƒ…ƒ‡ù‡ù‰oŠoŒåŒåŽ[�[‘ёѓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ï}!1Ðï}!1�¾!1L�¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1»ç!1(»ç!1Y("1¤Y("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1�Q"1€�Q"1+’"1ü+’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1+Ú"1´ +Ú"1‹þ"10 ‹þ"1Í+#1PÍ+#1c2#1,c2#1[‹#1˜[‹#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1í€&1Ìí€&1½R'1½R'1ee'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥B b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 f¦Äâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥B b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 f¦Äâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥B  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 f¦Äâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ©Ém� * bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1E€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1…ÈH‹ NÎ5 u  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ý ÝTTËË)B*B4¹4¹>2?2I©I©S T ^’^’hisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÝÝTTËË B B ¹ ¹00§§••  ƒƒúúq q"è"è$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIZJZLÑLÑNHOHQ¿Q¿S6T6V­V­X$Y$[›[›]^`‰`‰bcewewgîgîiejelÜlÜnNoNqÀqÀs2t2v¤v¤xy{ˆ{ˆ}ú}úl€l‚Þ‚Þ„P…P‡‡‰4Š4Œ¦Œ¦Ž�‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1k}!1Ðk}!1 ¾!1L ¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!17ç!1(7ç!1Õ'"1¤Õ'"1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1 Q"1€ Q"1§‘"1ü§‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1ñ4#1,ñ4#1�Œ#1˜�Œ#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p¿‹ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JP�þ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p¿‹ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JP�þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<p¿‹  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JP�þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 |œ@`1 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1”Ô€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1;»ˆÕU¢"d ¤  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙOOÅÅ);*;4±4±>'?'I›I›ST^ƒ^ƒh÷h÷rksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰ÿÿuuëëa a"×"×$M%M'Ã'Ã)9*9,¯,¯.%/%1›1›346‡6‡8ý8ý:s;s=é=é?_@_BÕBÕDKEKGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡‹‡‹‰ÿ‰ÿ‹sŒsŽçŽç�[‘[“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ï}!1Ðï}!1�¾!1L�¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1Áç!1(Áç!1_("1¤_("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1å4#1,å4#1ß‹#1˜ß‹#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦D b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 g§Åä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦D b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 g§Åä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦D  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 g§Åä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ©É�­Bb bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ˆÈ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1‘ÔT—ZÚ¸ ø  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ù ÙNNÃÃ)8*84­4­>$?$I™I™ST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÙÙNNÃà 8 8 ­ ­""——  ��öökkààU U"Ê"Ê$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1q}!1Ðq}!1¾!1L¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1Cç!1(Cç!1á'"1¤á'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1Q"1€Q"1³‘"1ü³‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1]2#1,]2#1KŒ#1˜KŒ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1ù€&1Ìù€&1ÉR'1ÉR'1qe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÂŽ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 &f9yÉ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÂŽ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 &f9yÉ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÂŽ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 &f9yÉ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �¯Jj.N bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÑQ¢"só `  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>-?-I¢I¢ST^Œ^Œhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbçbçd\e\gÑgÑiFjFl»l»n0o0q¥q¥stv�v�xy{y{y}î}îc€c‚؂؄M…M‡‡‰7Š7Œ¬Œ¬Ž!�!‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ã}!1Ðã}!1�¾!1L�¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1µç!1(µç!1S("1¤S("1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1‡Q"1€‡Q"1%’"1ü%’"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1%Ú"1´ %Ú"1…þ"10 …þ"1Ç+#1PÇ+#1]2#1,]2#1©‹#1˜©‹#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ÿ€&1Ìÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš7 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O�È›Û bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš7 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O�È›Û bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš7  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O�È›Û  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ©Éf†ü bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÉI’[Û€ À  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ä ä^^ÔÔ)J*J4À4À>6?6IªIªST^’^’hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjää^^ØØ R R Ì ÌFFÀÀ::´´..¨¨"" œ œ"#%Œ%Œ'(*x*x,î,î.d/d1Ú1Ú3P4P6Æ6Æ8<9<;²;²=(>(@ž@žBCEŠEŠGHJvJvLêLêN^O^QÒQÒSFTFVºVºX.Y.[¢[¢]^`Š`ŠbþbþdrergægæiZjZlÎlÎnBoBq¶q¶s*t*vžvžxy{†{†}ú}ún€n‚â‚â„V…V‡ʇʉ>Š>Œ²Œ²Ž&�&‘š‘š“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1q}!1Ðq}!1¾!1L¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1=ç!1(=ç!1Û'"1¤Û'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1Q"1€Q"1­‘"1ü­‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1mÚ"1´ mÚ"1Íþ"10 Íþ"1,#1P,#1ñ4#1,ñ4#1ÛŒ#1˜ÛŒ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÀŒ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 %e9yÊ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÀŒ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 %e9yÊ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÀŒ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 %e9yÊ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 o�ûAa bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1‘Ñ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1DÄ•æf·7± ñ  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÅÅ):*:4¯4¯>$?$I—I—S T ^}^}hðhðrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=ß=ß?T@TBÉBÉD>E>G³G³I(J(L›L›NOQ�Q�SôSôUgVgXÚXÚZM[M]À]À_3`3b¦b¦degŒgŒiÿiÿkrlrnånåpXqXsËsËu>v>x±x±z${$}—}— € ‚}‚}„ð„ð†c‡c‰Ö‰Ö‹IŒI޼޼�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1é}!1Ðé}!1‡¾!1L‡¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1µç!1(µç!1S("1¤S("1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1‡Q"1€‡Q"1%’"1ü%’"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1ë4#1,ë4#13Œ#1˜3Œ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›9 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 NŽÇœÜ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›9 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 NŽÇœÜ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›9  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 NŽÇœÜ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ©É†¦0P bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1‰É€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1$¤ím¶6ÿ I  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àXXÌÌ)@*@4´4´>(?(IšIšS T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  �� € €"ø"ø$l%l'à'à)T*T,È,È.Œ>ްް�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1w}!1Ðw}!1¾!1L¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Cç!1(Cç!1á'"1¤á'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1Q"1€Q"1³‘"1ü³‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1=Ú"1´ =Ú"1�þ"10 �þ"1ß+#1Pß+#1c2#1,c2#1ÿŒ#1˜ÿŒ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1Û€&1ÌÛ€&1«R'1«R'1Se'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÃ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @:zÍ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÃ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @:zÍ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÃ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @:zÍ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‚¢©É4 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€Ü\¸8”Ð  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á áWWÍÍ)C*C4¹4¹>-?-I¡I¡ST^‰^‰hýhýrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±''�� ‰ ‰"ÿ"ÿ$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@…@…BùBùDmEmGáGáIUJULÉLÉN=O=Q±Q±S%T%V™V™X Y [�[�]õ]õ_i`ibÝbÝdQeQgÅgÅi9j9l­l­n!o!q•q•s t v}v}xñxñze{e}Ù}ÙM€M‚Á‚Á„5…5‡©‡©‰ŠŒ�Œ�Ž�‘u‘u“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1Ý}!1ÐÝ}!1{¾!1L{¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1¯ç!1(¯ç!1M("1¤M("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1�Q"1€�Q"1’"1ü’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#1c2#1,c2#1]Œ#1˜]Œ#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�,ø b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R’Ë »û bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�,ø b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R’Ë »û bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�,ø  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R’Ë »û  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 _îSs bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÔT¨(|ü0 p  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝRRÇÇ)<*<4±4±>$?$I—I—S T ^}^}hðhðrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��wwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?P@PBÃBÃD6E6G©G©IJL�L�NOQuQuSèSèU[V[XÎXÎZA[A]´]´_'`'bšbšd e g€g€ióiókflfnÙnÙpLqLs¿s¿u2v2x¥x¥z{}‹}‹þþ�q‚q„ä„ä†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1q}!1Ðq}!1¾!1L¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Cç!1(Cç!1á'"1¤á'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1Q"1€Q"1³‘"1ü³‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1Ú"1´ Ú"1sþ"10 sþ"1µ+#1Pµ+#1ý4#1,ý4#1›�#1˜›�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1i€&1Ìi€&19R'19R'1ád'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÁ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @:zÎ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÁ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @:zÎ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÁ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @:zÎ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 b‚©Éå bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1œÜ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1pðLÌ(¨„l¬ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßSSÇÇ);*;4¯4¯>!?!I“I“ST^w^whéhér[s[} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ‡‡ûûo o"ã"ã$W%W'Ë'Ë)?*?,³,³.'/'1›1›346ƒ6ƒ8÷8÷:k;k=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b�b�dÿdÿfqgqiãiãkUlUnÇnÇp9q9s«s«uvx�x�z{}s}såå�W‚W„ɄɆ;‡;‰¬‰¬‹ŒŽŽŽŽ�ÿ�ÿ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1ã}!1Ðã}!1�¾!1L�¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1¯ç!1(¯ç!1M("1¤M("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1�Q"1€�Q"1’"1ü’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1÷4#1,÷4#1óŒ#1˜óŒ#1Ÿ$1$Ÿ$1wò$1èwò$1±Þ%1ì±Þ%1!€&1Ì!€&1ñQ'1ñQ'1™d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�.ú b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘Ê »û bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�.ú b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘Ê »û bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�.ú  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘Ê »û  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ŸDd> bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1”Ô€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1PÐ$¤øxÌLÄ  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝPPÃÃ)6*64©4©>?I‹I‹SüSü]m^mhÞhÞrOsO} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��ssææY Y"Ì"Ì$?%?'²'²)%*%,˜,˜. / 1~1~3ñ3ñ5d6d8×8×:J;J=½=½?.@.BŸBŸDEG�G�IòIòKcLcNÔNÔPEQES¶S¶U'V'X˜X˜Z [ ]z]z_ë_ëa\b\dÍdÍf>g>i¯i¯k l n‘n‘pqssssuäuäwUxUzÆzÆ|7}7¨¨�‚„ЄІû†ûˆk‰k‹Û‹Û�KŽK�»�»’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1}}!1Ð}}!1¾!1L¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Iç!1(Iç!1ç'"1¤ç'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1Q"1€Q"1¹‘"1ü¹‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1+Ú"1´ +Ú"1‹þ"10 ‹þ"1Í+#1PÍ+#1i2#1,i2#1³�#1˜³�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%15�&1Ì5�&1S'1S'1­e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÄ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [9yÏ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÄ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [9yÏ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÄ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [9yÏ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ©Éu•ú bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€çgÎNµ5€À €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg à àYYÒÒ)K*K4Ä4Ä>=?=I´I´S)T)^œ^œhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggààYYÒÒ K K Ä Ä==¶¶//¨¨!!šš Œ Œ"#%~%~'÷'÷)p*p,é,é.b/b1Û1Û3T4T6Í6Í8F9F;¿;¿=8>8@±@±B*C*E£E£GHJ•J•L M OƒOƒQúQúSqTqVèVèX_Y_[Ö[Ö]M^M`Ä`Äb9c9e®e®g#h#j˜j˜l m o€o€qóqósftfvÙvÙxLyL{¿{¿}2~2€¥€¥‚ƒ…‹…‹‡þ‡þ‰pŠpŒâŒâŽT�T‘ƑƓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1×}!1Ð×}!1u¾!1Lu¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1©ç!1(©ç!1G("1¤G("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1{Q"1€{Q"1’"1ü’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#1i2#1,i2#1�#1˜�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1Õ€&1ÌÕ€&1¥R'1¥R'1Me'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„!í b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•Φæ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„!í b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•Φæ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„!í  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•Î¦æ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 XxàNn bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ß_¾>�à  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âWWÌÌ)A*A4¶4¶>,?,I I ST^ˆ^ˆhühürrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢ŒŒ v v"ë"ë$`%`'Õ'Õ)J*J,¿,¿.4/41©1©346“6“89;};}=ò=ò?g@gBÜBÜDQEQGÆGÆI1?1I¦I¦ST^Š^Šhûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIYJYLÎLÎNCOCQ¸Q¸S-T-V¢V¢XY[Œ[Œ]^`v`vbébéd\e\gÏgÏiBjBlµlµn&o&q—q—stvyvyxêxêz[{[}Ì}Ì=€=‚®‚®„…‡‘‡‘‰ŠŒsŒsŽäŽä�U‘U“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1×}!1Ð×}!1u¾!1Lu¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1©ç!1(©ç!1G("1¤G("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1{Q"1€{Q"1’"1ü’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1 5#1, 5#1³�#1˜³�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…#ï b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T”Í §ç bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…#ï b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T”Í §ç bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…#ï  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T”Í §ç  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x˜ ,Uu bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ÿß€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1|ü[Û:º™¿ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞSSÈÈ)=*=4²4²>'?'IšIšS T ^~^~hðhðrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽxxííb b"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¤b¤degˆgˆiúiúklllnÞnÞpPqPsÂsÂu4v4x¦x¦z{}Š}Šüü�n‚n„à„à†T‡T‰ljNj:Œ:Ž­Ž­� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1ã}!1�ã}!1�¾!1�¾!1­Â!1$­Â!1¯ç!1�¯ç!1M("1M("1,"1$,"1�Q"1��Q"1’"1’"1Q–"1$Q–"1çØ"1�çØ"1Gý"1üGý"1)0#1 )0#1õ~#1| õ~#1ó$1Ü ó$1Ëò$1  Ëò$1ß%1¤ ß%1Y�&1„ Y�&1)S'1È)S'1Ñe'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qQ…Ö¢'ó b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qQ…Ö¢'ó b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qQ…Ö¢'ó b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1©É  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx ð ðhhàà)X*X4Ð4Ð>F?FI¼I¼S2T2^¨^¨his”s”} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðxxððhhàà X X Ð ÐHHÀÀ88°°((   � �"#%€%€'ø'ø)p*p,è,è.`/`1Ø1Ø3P4P6È6È8@9@;¸;¸=0>0@¨@¨B C E˜E˜GHJ†J†LüLüNrOrQèQèS^T^VÔVÔXJYJ[À[À]6^6`¬`¬b"c"e˜e˜ghj„j„lúlúnpopqæqæs\t\vÒvÒxHyH{¾{¾}4~4€ª€ª‚ ƒ …–…–‡ ˆ ЂЂŒøŒøŽn�n‘ä‘ä“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ñ}!1�Ñ}!1o¾!1 o¾!1wÅ!1wÅ!1£ç!1€£ç!1A("1üA("1I/"1I/"1uQ"1puQ"1’"1ì’"1™"1ô™"1…Ú"1`…Ú"1åþ"1Ìåþ"1)0#1ì)0#1õ~#1L õ~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1 �&1T  �&1ÛR'1˜ÛR'1ƒe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dxá b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ç€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dxá b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ç€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Dxá b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ç €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ff  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss ê êaaØØ)O*O4Æ4Æ>=?=I´I´S+T+^�^�his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssêêaaØØ O O Æ Æ==´´++¢¢�� ~ ~"õ"õ$l%l'ã'ã)Z*Z,Ñ,Ñ.H/H1¿1¿36466­6­8$9$;›;›=>@‰@‰BCEwEwGîGîIeJeLÜLÜNSOSQÊQÊSATAV¸V¸X/Y/[¦[¦]^`”`”b c e‚e‚gùgùipjplçlçnYoYqËqËs=t=v¯v¯x!y!{“{“}~€w€w‚é‚é„[…[‡͇͉?Š?Œ±Œ±Ž#�#‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1ã}!1�ã}!1�¾!1�¾!1­Â!1$­Â!1µç!1�µç!1S("1S("1,"1$,"1‡Q"1�‡Q"1%’"1%’"1Q–"1$Q–"1çØ"1�çØ"1Gý"1üGý"1)0#1 )0#1õ~#1| õ~#1Û$1Ü Û$1³ò$1  ³ò$1íÞ%1¤ íÞ%1á€&1„ á€&1±R'1ȱR'1Ye'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rOƒÒž*ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rOƒÒž*ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rOƒÒž*ö b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1©É  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ì ìbbØØ)N*N4Ä4Ä>8?8I¬I¬S T ^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvììbbØØ N N Ä Ä::°°&&œœˆˆþþt t"ê"ê$`%`'Ö'Ö)L*L,Â,Â.8/81®1®3$4$6š6š89;†;†=ü=ü?r@rBèBèD^E^GÔGÔIHJHL¼L¼N0O0Q¤Q¤STVŒVŒXY[t[t]è]è_\`\bÐbÐdDeDg¸g¸i,j,l l noqˆqˆsüsüupvpxäxäzX{X}Ì}Ì@€@‚´‚´„(…(‡œ‡œ‰ŠŒ„Œ„ŽøŽø�l‘l“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ñ}!1�Ñ}!1o¾!1 o¾!1wÅ!1wÅ!1£ç!1€£ç!1A("1üA("1I/"1I/"1uQ"1puQ"1’"1ì’"1™"1ô™"1‹Ú"1`‹Ú"1ëþ"1Ìëþ"1)0#1ì)0#1õ~#1L õ~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1±€&1T ±€&1�R'1˜�R'1)e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cwá b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ç€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cwá b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ç€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cwá b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ç €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Oo  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss è è]]ÒÒ)G*G4¼4¼>1?1I¦I¦ST^�^�hÿhÿrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssèè]]ÒÒ G G ¼ ¼11¦¦��zzïïd d"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I)J)LžLžNOQˆQˆSýSýUrVrXçXçZ\[\]Ñ]Ñ_F`Fb»b»d0e0g¥g¥ijl�l�noqsqssåsåuWvWxÉxÉz;{;}­}­€‚‘‚‘„…‡u‡u‰ç‰ç‹YŒYŽËŽË�=‘=“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1�}!1��}!1-¾!1 -¾!1ÅÂ!1ÅÂ!1[ç!1€[ç!1ù'"1üù'"1—,"1—,"1-Q"1p-Q"1Ë‘"1ìË‘"1i–"1ôi–"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1�&1T �&1çR'1˜çR'1�e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ÓŸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1[€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ÓŸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1[€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ÓŸ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1[ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1©É  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ë ëbbÙÙ)P*P4Ç4Ç>>µµ,,££‘‘  "ö"ö$m%m'ä'ä)[*[,Ò,Ò.I/I1À1À37476®6®8%9%;œ;œ=>@Š@ŠBCExExGïGïIdJdLÙLÙNNONQÃQÃS8T8V­V­X"Y"[—[—] ^ `�`�böbödkekgàgàiUjUlÊlÊn?o?q´q´s)t)vžvžxy{ˆ{ˆ}ý}ýr€r‚ç‚ç„\…\‡чщFŠFŒ»Œ»Ž0�0‘¥‘¥“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1×}!1�×}!1u¾!1 u¾!1MÅ!1MÅ!1©ç!1€©ç!1G("1üG("1/"1/"1{Q"1p{Q"1’"1ì’"1ñ˜"1ôñ˜"1yÚ"1`yÚ"1Ùþ"1ÌÙþ"1#0#1ì#0#1ï~#1L ï~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ê b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¦æ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ê b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¦æ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ê b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¦æ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1?_  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss í íggÝÝ)S*S4É4É>=?=I±I±S%T%^™^™h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssííggáá [ [ Õ ÕOOÉÉCC½½77±±++ ¥ ¥"#%•%•' ( *�*�,÷,÷.m/m1ã1ã3Y4Y6Ï6Ï8E9E;»;»=1>1@§@§BCE“E“G H J}J}LñLñNeOeQÙQÙSMTMVÁVÁX5Y5[©[©]^`‘`‘bceyeygígíiajalÕlÕnIoIq½q½s1t1v¥v¥xy{�{�}~€u€u‚é‚é„]…]‡чщEŠEŒ¹Œ¹Ž-�-‘¡‘¡“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1‰}!1�‰}!1'¾!1 '¾!1ÅÂ!1ÅÂ!1[ç!1€[ç!1ù'"1üù'"1—,"1—,"1-Q"1p-Q"1Ë‘"1ìË‘"1i–"1ôi–"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Ï› b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Z€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Ï› b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Z€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Ï› b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Z €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«Ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss é é__ÔÔ)I*I4¾4¾>1?1I¤I¤ST^Š^Šhýhýrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðsséé__ÕÕ K K Á Á77­­##™™……ûûq q"ç"ç$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI5J5L¨L¨NOQŽQŽSTVtVtXçXçZZ[Z]Í]Í_@`@b³b³d&e&g™g™i j llnònòpeqesØsØuKvKx¾x¾z1{1}¤}¤€‚ЂЄý„ý†p‡p‰ã‰ã‹VŒVŽÉŽÉ�<‘<“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1×}!1�×}!1u¾!1 u¾!1SÅ!1SÅ!1©ç!1€©ç!1G("1üG("1%/"1%/"1{Q"1p{Q"1’"1ì’"1÷˜"1ô÷˜"1‹Ú"1`‹Ú"1ëþ"1Ìëþ"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€ê b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1§ç€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€ê b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1§ç€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€ê b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1§ç €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Pp  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss ë ëcc××)K*K4¿4¿>1?1I£I£ST^‡^‡hùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssëëccÛÛ S S Ë ËCC»»33««##›› ‹ ‹"#%w%w'ë'ë)_*_,Ó,Ó.G/G1»1»3/4/6£6£89;‹;‹=ÿ=ÿ?s@sBçBçD[E[GÏGÏIAJAL³L³N%O%Q—Q—S T V{V{XíXíZ_[_]Ñ]Ñ_C`Cbµbµd'e'g™g™i j l}l}nïnïpaqasÓsÓuEvEx·x·z){)}›}› € ‚‚„ñ„ñ†c‡c‰Õ‰Õ‹GŒG޹޹�+‘+“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1‰}!1�‰}!1'¾!1 '¾!1­Â!1­Â!1Uç!1€Uç!1ó'"1üó'"1,"1,"1'Q"1p'Q"1Å‘"1ìÅ‘"1Q–"1ôQ–"1yÚ"1`yÚ"1Ùþ"1ÌÙþ"1#0#1ì#0#1ï~#1L ï~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1í€&1T í€&1½R'1˜½R'1ee'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KМ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KМ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KМ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Dd  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt î îddÚÚ)P*P4Ä4Ä>8?8I¬I¬S T ^”^”hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttîîhhââ \ \ Ö ÖPPÊÊDD¾¾44ªª  – –" # %‚%‚'ø'ø)n*n,ä,ä.Z/Z1Ð1Ð3F4F6¼6¼8292;¨;¨=>@�@�BCExExGìGìI`J`LÔLÔNHOHQ¼Q¼S0T0V¤V¤XY[Œ[Œ]^`t`tbèbèd\e\gÐgÐiDjDl¸l¸n,o,q q stvˆvˆxüxüzp{p}ä}äX€X‚̂̄@…@‡²‡²‰%Š%Œ˜Œ˜Ž � ‘~‘~“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Ý}!1�Ý}!1{¾!1 {¾!1wÅ!1wÅ!1¯ç!1€¯ç!1M("1üM("1I/"1I/"1�Q"1p�Q"1’"1ì’"1™"1ô™"1ÛØ"1`ÛØ"1;ý"1Ì;ý"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ'ó b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ï€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ'ó b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ï€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ'ó b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ï €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1©É  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll â âWWÌÌ)A*A4´4´>'?'IšIšS T ^€^€hóhórdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllââXXÎÎ D D º º00¦¦’’||ññf f"Û"Û$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=à=à?S@SBÆBÆD9E9G¬G¬IJL’L’NOQxQxSëSëU^V^XÑXÑZD[D]·]·_*`*b�b�degƒgƒiöiökilinÜnÜpOqOsÂsÂu5v5x¨x¨z{}Ž}Ž€‚t‚t„ç„ç†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1ƒ}!1�ƒ}!1!¾!1 !¾!1­Â!1­Â!1Uç!1€Uç!1ó'"1üó'"1,"1,"1'Q"1p'Q"1Å‘"1ìÅ‘"1Q–"1ôQ–"1Ú"1`Ú"1ßþ"1Ìßþ"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1u€&1T u€&1ER'1˜ER'1íd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}̘ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}̘ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}̘ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Mm  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ê ê^^ÒÒ)F*F4¸4¸>*?*IœIœST^€^€hòhòrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrêêbbÚÚ R R Ê ÊBBºº22ªª’’ z z"î"î$b%b'Ö'Ö)J*J,¾,¾.2/21¦1¦346Ž6Ž89;v;v=è=è?Z@ZBÌBÌD>E>G°G°I"J"L”L”NOQxQxSêSêU\V\XÎXÎZ@[@]²]²_$`$b–b–degzgziìiìk^l^nÐnÐpBqBs´s´u&v&x˜x˜z { }|}|îî�`‚`„Ò„Ò†B‡B‰³‰³‹$Œ$ޕޕ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Ý}!1�Ý}!1{¾!1 {¾!1qÅ!1qÅ!1¯ç!1€¯ç!1M("1üM("1C/"1C/"1�Q"1p�Q"1’"1ì’"1™"1ô™"1ÛØ"1`ÛØ"1;ý"1Ì;ý"1#0#1ì#0#1ï~#1L ï~#1¥$1¬ ¥$1}ò$1p }ò$1·Þ%1t ·Þ%1-€&1T -€&1ýQ'1˜ýQ'1¥d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰'ó b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1΀} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰'ó b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1΀} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰'ó b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Î €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1©É  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn ä äWWÊÊ)=*=4®4®>?I�I�ST^r^rhãhãrRsR} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnääZZÐÐ F F ¼ ¼22¨¨””zzíí` `"Ó"Ó$F%F'¹'¹),*,,Ÿ,Ÿ./1…1…3ø3ø5k6k8Þ8Þ:Q;Q=Â=Â?3@3B¤B¤DEG†G†I÷I÷KhLhNÙNÙPJQJS»S»U,V,X�X�Z[]]_ð_ðaabadÒdÒfCgCi´i´k%l%n–n–pqsxsxuéuéwZxZzËzË|<}<­­�‚„�„�†þ†þˆn‰n‹Þ‹Þ�NŽN�¾�¾’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1}}!1�}}!1¾!1 ¾!1ÑÂ!1ÑÂ!1Oç!1€Oç!1í'"1üí'"1£,"1£,"1!Q"1p!Q"1¿‘"1ì¿‘"1u–"1ôu–"1mÚ"1`mÚ"1Íþ"1ÌÍþ"10#1ì0#1é~#1L é~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%15�&1T 5�&1S'1˜S'1­e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Í™ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations12r€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Í™ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations12r€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Í™ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations12r €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1?_  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp é ébbÛÛ)T*T4Í4Í>D?DI»I»S0T0^£^£his‡s‡} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppéébbÛÛ T T Í ÍFF¿¿88±±**££ • •"#%‡%‡'(*y*y,ò,ò.k/k1ä1ä3]4]6Ö6Ö8O9O;È;È=A>A@º@ºB3C3E¬E¬G%H%JœJœLMOŠOŠQRTxTxVïVïXfYf[Ý[Ý]T^T`Ë`Ëb@c@eµeµg*h*jŸjŸlmo‡o‡qúqúsmtmvàvàxSyS{Æ{Æ}9~9€¬€¬‚ƒ…’…’‡ˆŠuŠuŒçŒçŽY�Y‘ˑ˓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ã}!1�ã}!1�¾!1 �¾!1;Å!1;Å!1µç!1€µç!1S("1üS("1 /"1 /"1‡Q"1p‡Q"1%’"1ì%’"1ߘ"1ôߘ"1ÛØ"1`ÛØ"1;ý"1Ì;ý"10#1ì0#1é~#1L é~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“0ü b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1—×€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“0ü b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1—×€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“0ü b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1—× €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1©É  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk ã ãXXÍÍ)B*B4·4·>+?+IŸIŸST^‡^‡hûhûroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkãã[[ÓÓ K K à Ã;;³³++££�� w w"ì"ì$a%a'Ö'Ö)K*K,À,À.5/51ª1ª346”6”8 9 ;~;~=ó=ó?h@hBÝBÝDRERGÇGÇI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÃbÃd7e7g«g«ijl“l“noq{q{sïsïucvcx×x×zK{K}¿}¿3€3‚§‚§„…‡�‡�‰ŠŒyŒyŽîŽî�c‘c“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1}}!1�}}!1¾!1 ¾!1×Â!1×Â!1Iç!1€Iç!1ç'"1üç'"1©,"1©,"1Q"1pQ"1¹‘"1칑"1{–"1ô{–"1…Ú"1`…Ú"1åþ"1Ìåþ"10#1ì0#1é~#1L é~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1À&1T À&1“R'1˜“R'1;e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÉ• b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations13s€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÉ• b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations13s€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÉ• b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations13s €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Tt  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn å å\\ÓÓ)J*J4Á4Á>6?6I«I«ST^�^�hispsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnåå\\ÓÓ J J Á Á88¯¯&&��‹‹ y y"ð"ð$g%g'Þ'Þ)U*U,Ì,Ì.C/C1º1º31416¨6¨89;–;–= > @„@„BûBûDrErGéGéI^J^LÓLÓNHOHQ½Q½S2T2V§V§XY[‘[‘]^`{`{bîbîdaeagÔgÔiGjGlºlºn+o+qœqœs t v~v~xïxïz`{`}Ñ}ÑB€B‚³‚³„$…$‡”‡”‰ŠŒvŒvŽçŽç�X‘X“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ã}!1�ã}!1�¾!1 �¾!1;Å!1;Å!1µç!1€µç!1S("1üS("1 /"1 /"1‡Q"1p‡Q"1%’"1ì%’"1ߘ"1ôߘ"1ÛØ"1`ÛØ"1;ý"1Ì;ý"10#1ì0#1é~#1L é~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’0ü b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1—×€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’0ü b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1—×€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’0ü b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1—× €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1©É  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk á áVVËË)@*@4µ4µ>(?(I›I›S T ^^hñhñrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkááWWÍÍ C C ¹ ¹//¥¥‘‘{{ððe e"Ú"Ú$O%O'Ä'Ä)9*9,®,®.#/#1˜1˜3 4 6‚6‚8÷8÷:l;l=á=á?V@VBËBËD@E@GµGµI(J(L›L›NOQ�Q�SôSôUgVgXÚXÚZM[M]À]À_3`3b¥b¥deg‰g‰iûiûkmlmnßnßpQqQsÃsÃu5v5x§x§z{}‹}‹ýý�o‚o„á„á†S‡S‰ƉƋ9Œ9ެެ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1w}!1�w}!1¾!1 ¾!1­Â!1­Â!1Iç!1€Iç!1ç'"1üç'"1,"1,"1Q"1pQ"1¹‘"1칑"1Q–"1ôQ–"1gÚ"1`gÚ"1Çþ"1ÌÇþ"10#1ì0#1ã~#1L ã~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1;�&1T ;�&1 S'1˜ S'1³e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÊ– b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÊ– b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÊ– b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations18X  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ä ä\\ÔÔ)L*L4Ä4Ä>:?:I°I°S&T&^œ^œhisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllää\\ÔÔ L L Ä Ä<<´´,,¤¤””  „ „"ü"ü$t%t'ì'ì)d*d,Ü,Ü.T/T1Ì1Ì3D4D6¼6¼8494;¬;¬=$>$@œ@œBCEŒEŒGHJzJzLðLðNfOfQÜQÜSRTRVÈVÈX>Y>[´[´]*^*` ` bceŒeŒghjxjxlîlîndodqÚqÚsPtPvÆvÆx1?1I¨I¨ST^‘^‘hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIYJYLÐLÐNGOGQ¾Q¾S5T5V¬V¬X#Y#[š[š]^`ˆ`ˆbÿbÿdvevgígíidjdlÛlÛnMoMq¿q¿s1t1v£v£xy{‡{‡}ù}ùk€k‚݂݄O…O‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1w}!1�w}!1¾!1 ¾!1­Â!1­Â!1Cç!1€Cç!1á'"1üá'"1,"1,"1Q"1pQ"1³‘"1쳑"1Q–"1ôQ–"1Ú"1`Ú"1ßþ"1Ìßþ"10#1ì0#1ã~#1L ã~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1À&1T À&1“R'1˜“R'1;e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwÆ’ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwÆ’ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwÆ’ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Uu  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj à àVVÌÌ)B*B4¸4¸>,?,I I ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjààVVÌÌ B B ¸ ¸..¤¤��||òòh h"Þ"Þ$T%T'Ê'Ê)@*@,¶,¶.,/,1¢1¢346Ž6Ž89;z;z=ð=ð?f@fBÜBÜDRERGÈGÈI%?%IšIšST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÜÜQQÆÆ ; ; ° °%%šš„„ùùnnããX X"Í"Í$B%B'·'·),*,,¡,¡./1‹1‹346u6u8ê8ê:_;_=Ô=Ô?I@IB¾B¾D3E3G¨G¨IJL’L’NOQ|Q|SñSñUfVfXÛXÛZP[P]Å]Å_:`:b¯b¯d$e$g™g™ijlƒlƒnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1q}!1�q}!1¾!1 ¾!1ËÂ!1ËÂ!1Cç!1€Cç!1á'"1üá'"1�,"1�,"1Q"1pQ"1³‘"1쳑"1o–"1ôo–"1[Ú"1`[Ú"1»þ"1Ì»þ"10#1ì0#1ã~#1L ã~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1ù€&1T ù€&1ÉR'1˜ÉR'1qe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÇ“ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1/o€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÇ“ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1/o€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÇ“ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1/o €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations13S  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>0?0I¥I¥ST^�^�hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubêbêd_e_gÔgÔiIjIl¾l¾n3o3q¨q¨stv’v’xy{|{|}ñ}ñf€f‚Û‚Û„P…P‡Ňʼn:Š:Œ¯Œ¯Ž$�$‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1ï}!1�ï}!1�¾!1 �¾!1/Å!1/Å!1»ç!1€»ç!1Y("1üY("1/"1/"1�Q"1p�Q"1+’"1ì+’"1Ó˜"1ôÓ˜"1ÏØ"1`ÏØ"1/ý"1Ì/ý"10#1ì0#1ã~#1L ã~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1ó€&1T ó€&1ÃR'1˜ÃR'1ke'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥B b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1—×€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥B b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1—×€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥B b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1—× €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1©É  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á á[[ÑÑ)G*G4½4½>1?1I¥I¥ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±++¥¥ ™ ™"#%‰%‰'ÿ'ÿ)u*u,ë,ë.a/a1×1×3M4M6Ã6Ã8999;¯;¯=%>%@›@›BCE‡E‡GýGýIqJqLåLåNYOYQÍQÍSATAVµVµX)Y)[�[�]^`…`…bùbùdmemgágáiUjUlÉlÉn=o=q±q±s%t%v™v™x y {�{�}õ}õi€i‚݂݄Q…Q‡Ňʼn9Š9Œ­Œ­Ž!�!‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1k}!1�k}!1 ¾!1  ¾!1ËÂ!1ËÂ!1=ç!1€=ç!1Û'"1üÛ'"1�,"1�,"1Q"1pQ"1­‘"1ì­‘"1o–"1ôo–"1Ú"1`Ú"1ßþ"1Ìßþ"10#1ì0#1ã~#1L ã~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÃ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations10p€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÃ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations10p€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÃ� b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations10p €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Tt  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝSSÈÈ)=*=4²4²>%?%I˜I˜S T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��yyïïe e"Û"Û$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnænæpYqYsÌsÌu?v?x²x²z%{%}˜}˜ € ‚~‚~„ñ„ñ†d‡d‰׉׋JŒJ޽޽�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1ï}!1�ï}!1�¾!1 �¾!1/Å!1/Å!1»ç!1€»ç!1Y("1üY("1/"1/"1�Q"1p�Q"1+’"1ì+’"1Ó˜"1ôÓ˜"1ÏØ"1`ÏØ"1/ý"1Ì/ý"10#1ì0#1ã~#1L ã~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤B b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1—×€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤B b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1—×€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤B b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1—× €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1©É  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßWWËË)?*?4³4³>%?%I—I—S T ^{^{híhír_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ��  "÷"÷$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnãnãpUqUsÇsÇu9v9x«x«z{}�}�€‚s‚s„å„å†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1k}!1�k}!1 ¾!1  ¾!1¡Â!1¡Â!1=ç!1€=ç!1Û'"1üÛ'"1s,"1s,"1Q"1pQ"1­‘"1ì­‘"1E–"1ôE–"1ÏØ"1`ÏØ"1/ý"1Ì/ý"10#1ì0#1Ý~#1L Ý~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1Ï€&1T Ï€&1ŸR'1˜ŸR'1Ge'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÄ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÄ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÄ� b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1©É  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh â âXXÎÎ)D*D4¸4¸>,?,I I ST^Š^Šhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²((žž Š Š"#%v%v'ì'ì)b*b,Ø,Ø.N/N1Ä1Ä3:4:6°6°8&9&;œ;œ=>@„@„BøBøDlElGàGàITJTLÈLÈN?IŽIŽS T ^~^~hñhñrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††ûûppååZ Z"Ï"Ï$D%D'¹'¹).*.,£,£./1�1�346w6w8ì8ì:a;a=Ô=Ô?G@GBºBºD-E-G G IJL†L†NùNùPlQlSßSßURVRXÅXÅZ8[8]«]«_`b›b›deg�g�iôiôkglgnÚnÚpMqMsÀsÀu3v3x¦x¦z{}Œ}Œÿÿ�r‚r„å„å†V‡V‰ȉȋ:Œ:ެެ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1e}!1�e}!1¾!1 ¾!1¡Â!1¡Â!17ç!1€7ç!1Õ'"1üÕ'"1s,"1s,"1 Q"1p Q"1§‘"1ì§‘"1E–"1ôE–"1ÏØ"1`ÏØ"1/ý"1Ì/ý"10#1ì0#1Ý~#1L Ý~#1Ÿ$1¬ Ÿ$1wò$1p wò$1±Þ%1t ±Þ%1]€&1T ]€&1-R'1˜-R'1Õd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÀŒ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÀŒ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÀŒ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1©É  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞRRÆÆ):*:4¬4¬>?I�I�ST^x^xhêhêrZsZ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žž††úún n"â"â$V%V'Ê'Ê)>*>,²,².&/&1š1š346‚6‚8ö8ö:j;j=Ü=Ü?N@NBÀBÀD2E2G¤G¤IJLˆLˆNúNúPlQlSÞSÞUPVPXÂXÂZ4[4]¦]¦_`bŽbŽdegrgriäiäkVlVnÈnÈp:q:s¬s¬uvx�x�z{}t}tææ�X‚X„ʄʆ:‡:‰«‰«‹ŒŽ�Ž��þ�þ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1ï}!1�ï}!1�¾!1 �¾!1_Å!1_Å!1Áç!1€Áç!1_("1ü_("11/"11/"1“Q"1p“Q"11’"1ì1’"1™"1ô™"11Ú"1`1Ú"1‘þ"1Ì‘þ"10#1ì0#1Ý~#1L Ý~#1“$1¬ “$1»ð$1p »ð$1™À%1t ™À%1€&1T €&1 P'1˜ P'1³b'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­K b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1È€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­K b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1È€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­K b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1È €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1/+  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØKK¾¾)1*14¢4¢>?I„I„SýSý]n^nhßhßrNsN} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆûûnnááT T"Ç"Ç$:%:'­'­) * ,“,“./1y1y3ì3ì5_6_8Ò8Ò:E;E=¶=¶?'@'B˜B˜D E GzGzIëIëK\L\NÍNÍP>Q>S¯S¯U V X‘X‘Z[]s]s_ä_äa]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L%i!1e}!1�e}!1¾!1 ¾!1¿Â!1¿Â!11ç!1€1ç!1Ï'"1üÏ'"1‘,"1‘,"1Q"1pQ"1¡‘"1ì¡‘"1c–"1ôc–"1[Ú"1`[Ú"1»þ"1Ô»þ"1 0#1ô 0#1×~#1T ×~#1É$1´ É$1ñð$1x ñð$1ÏÀ%1| ÏÀ%1#�&1\ #�&1Q'1 Q'1Ác'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<pÁ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1,l€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<pÁ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1,l€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<pÁ� b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1,l €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1µ�4X  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ý ÝVVÑÑ)J*J4Ã4Ã>:?:I±I±S&T&^¥^¥his‰s‰} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÝÝVVÏÏ H H Á Á::³³,,¥¥—— ‰ ‰"#%}%}'ö'ö)o*o,è,è.a/a1Ú1Ú3S4S6Ì6Ì8E9E;¾;¾=7>7@°@°B)C)E¢E¢GHJ’J’L M O€O€Q÷Q÷SnTnVåVåX\Y\[Ó[Ó]J^J`Á`Áb6c6e«e«g h j•j•l m o‰o‰qüqüsotovâvâxUyU{È{È};~;€®€®‚!ƒ!…”…”‡ˆŠwŠwŒéŒéŽ[�[‘͓͑ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L%i!1õ}!1�õ}!1“¾!1 “¾!1)Å!1)Å!1Çç!1€Çç!1e("1üe("1û."1û."1™Q"1p™Q"17’"1ì7’"1͘"1ô͘"1¥Ø"1`¥Ø"1ý"1Ìý"1 0#1ì 0#1×~#1L ×~#1½$1¬ ½$1åð$1p åð$1ÃÀ%1t ÃÀ%1À&1T À&1¹P'1˜¹P'1ac'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·T  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1—×€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·T  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1—×€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·T  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1—× €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1©¥  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ × ×LLËË)@*@4µ4µ>)?)I�I�ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__××OOÇÇ ? ? · ·//§§——  ��öök k"à"à$_%_'Ô'Ô)I*I,¾,¾.3/31¨1¨346’6’89;|;|=ñ=ñ?f@fBÛBÛDPEPGÅGÅI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÍbÍdAeAgµgµi)j)l�l�noq…q…sùsùumvmxáxázU{U}É}É=€=‚±‚±„%…%‡™‡™‰ŠŒƒŒƒŽøŽø�m‘m“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L%i!1_}!1�_}!1ý½!1 ý½!1¿Â!1¿Â!11ç!1€1ç!1Ï'"1üÏ'"1‘,"1‘,"1Q"1pQ"1¡‘"1ì¡‘"1c–"1ôc–"1Ú"1`Ú"1ßþ"1Ôßþ"1 0#1ô 0#1×~#1T ×~#1·$1´ ·$1ßð$1x ßð$1½À%1| ½À%1·€&1\ ·€&1­P'1 ­P'1Uc'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n½‰ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1-m€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n½‰ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1-m€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:n½‰ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1-m €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1µ�Y}  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPËË)B*B4¹4¹>.?.I£I£ST^“^“histst} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbæbædYeYgÌgÌi?j?l²l²n/o/q q stv‚v‚xóxózd{d}Õ}ÕF€F‚·‚·„(…(‡˜‡˜‰ Š ŒzŒzŽëŽë�\‘\“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L%i!1õ}!1�õ}!1“¾!1 “¾!1)Å!1)Å!1Çç!1€Çç!1e("1üe("1û."1û."1™Q"1p™Q"17’"1ì7’"1͘"1ô͘"1¥Ø"1`¥Ø"1ý"1Ìý"1 0#1ì 0#1×~#1L ×~#1«$1¬ «$1Óð$1p Óð$1±À%1t ±À%1�€&1T �€&1wP'1˜wP'1c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶T  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1—×€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶T  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1—×€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶T  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1—× €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1©¥  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Õ ÕJJÇÇ)<*<4±4±>$?$I—I—ST^‡^‡hùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……úúooääY Y"Î"Î$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD4?4I«I«S"T"^ ^ his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð@@··..¥¥   ± ±((ŸŸ��{{òòi i"à"à$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãI\J\LÓLÓNJOJQÁQÁS8T8V¯V¯X&Y&[�[�]^`‹`‹bceyeygðgðigjglÞlÞn\o\qÎqÎs@t@v²v²x$y${–{–}~€z€z‚ì‚ì„^…^‡ЇЉBŠBŒ´Œ´Ž&�&‘˜‘˜“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxi!1~!1è~!1¥¾!1d¥¾!1MÅ!1lMÅ!1×Î!1è×Î!1 Ò!1´ Ò!1Óç!1@Óç!1q("1¼q("1/"1Ä/"1©8"1@©8"1ß;"1 ß;"1¥Q"1˜¥Q"1C’"1C’"1ñ˜"1 ñ˜"1{¢"1˜ {¢"1±¥"1d ±¥"1óÒ"1ð óÒ"1sÚ"1Ì sÚ"1Óþ"1H Óþ"1,#1h,#1:#1D:#1Ãn#1ÐÃn#1=Ž#1ô=Ž#1;™#1 ;™#1É$1¬É$1¡ò$1p¡ò$1ÛÞ%1tÛÞ%1Uì%1TUì%1/�&1à/�&1ÿR'1$ÿR'1§e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1›Ïj6 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Cƒ©é¾þ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1›Ïj6 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Cƒ©é¾þ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1›Ïj6  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Cƒ©é¾þ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ˜¸\|`€ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1<|¸ø6vuµ´ô  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d<¼øx´4pð,¬èh¤$`àžÜ\—šòrXØÍM  – ¨ ( Ô T ’  P Ð  L ˆ È  F „ Ä %e bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSÍÍ)E*E4Ÿ4Ÿ>5?5I«I«S!T!^£^£his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=ÿ=ÿ?w@wBïBïDgEgGßGßIuJuLëLëNaOaQ×Q×SMTMVÃVÃX9Y9[¯[¯]%^%`›`›bce‡e‡gýgýisjslélénkokqáqásWtWvÍvÍxCyC{¹{¹}/~/€¥€¥‚ƒ…‘…‘‡ˆŠ}Š}ŒóŒóŽi�i‘ߑߓ  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„i!15}!15}!1Ó½!1|Ó½!1=Ã!1„=Ã!1sÆ!1ðsÆ!1ýÏ!1|ýÏ!13Ó!1H3Ó!1ç!1Ôç!1¥'"1P¥'"1-"1X-"1E0"1ÄE0"1Ï9"1PÏ9"1="1="1ÙP"1¨ÙP"1w‘"1$ w‘"1á–"1, á–"1š"1˜ š"1¡£"1$ ¡£"1צ"1ð צ"1Ô"1| Ô"1…Ú"1X…Ú"1åþ"1Ôåþ"1',#1ô',#1¯:#1Я:#1ko#1\ko#1ç�#1€ç�#1åš#1¬åš#1«$18«$1ƒò$1üƒò$1½Þ%1½Þ%17ì%1à7ì%1�€&1l�€&1]R'1°]R'1e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O l b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1«ë bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O l b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1«ë  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O l  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1«ë bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ­Í«Ër’ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1h¨P¶ö¶ö \œ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dhèPÐ8¸  ˆðpØXÀ@¦&Œ r ò d ä X Ø ) © î n > ¾ ³ 3 $ ¤  Š ð pGØ€À&fÌ  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10?? Ó ÓHHÅÅ):*:4¯4¯>&?&I›I›ST^Ž^Žhisrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð??´´))žž   § §‘‘{{ððeeÚÚO O"Ä"Ä$A%A'¶'¶)+*+, , ./1Š1Š3ÿ3ÿ5t6t8é8é:^;^=Ó=Ó?H@HB½B½D2E2G§G§IJL“L“NOQ}Q}SòSòUgVgXÜXÜZQ[Q]Æ]Æ_;`;b°b°d%e%gšgšijl„l„noqtqtsæsæuXvXxÊxÊz<{<}®}® € ‚’‚’„…‡v‡v‰è‰è‹ZŒZŽÌŽÌ�>‘>“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxi!1~!1è~!1¥¾!1d¥¾!1MÅ!1lMÅ!1×Î!1è×Î!1 Ò!1´ Ò!1Ùç!1@Ùç!1w("1¼w("1/"1Ä/"1©8"1@©8"1ß;"1 ß;"1«Q"1˜«Q"1I’"1I’"1ñ˜"1 ñ˜"1{¢"1˜ {¢"1±¥"1d ±¥"1óÒ"1ð óÒ"1Ú"1Ì Ú"1ßþ"1H ßþ"1!,#1h!,#1[:#1D[:#1o#1Ðo#1ÙŽ#1ôÙŽ#1×™#1 ×™#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1½€&1ཀ&1�R'1$�R'15e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐl8 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 D„ªê¿ÿ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐl8 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 D„ªê¿ÿ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐl8  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 D„ªê¿ÿ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¬Ì-Mhˆ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1G‡ÎS“Ø P  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dGÇŽÕUœcã*ªñq¸8}ýB‡ÌL ‘  V Ö ƒ   › ^ Þ à ` 9 ¹ § ç  ” . n ³ ó 8xÀ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×MMÇÇ)=*=4”4”>)?)I�I�ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=Ì=Ì?B@BB¸B¸D.E.G¤G¤I9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noq…q…sùsùumvmxáxázU{U}É}É=€=‚±‚±„%…%‡™‡™‰ Š Œ�Œ�ŽõŽõ�i‘i“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1_}!1è_}!1ý½!1dý½!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!11ç!1@1ç!1Ï'"1¼Ï'"1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1Q"1˜Q"1¡‘"1¡‘"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"1gÚ"1Ì gÚ"1Çþ"1H Çþ"1 ,#1h ,#1:#1D:#1½n#1нn#1½Œ#1ô½Œ#1»—#1 »—#1½$1¬½$1•ò$1p•ò$1ÏÞ%1tÏÞ%1Iì%1TIì%1ó€&1àó€&1ÃR'1$ÃR'1ke'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;oÂŽ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 8x;{È bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;oÂŽ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 8x;{È bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;oÂŽ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 8x;{È  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 «Ë�°Pp bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1;{¶ö1q.n¬ì  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d;»öv±1lì'§âb�XØ“ZÚÎN!¡‰ èhDį/ ÿ º : u õ 0 ° ë + f ¦ á ! \ œ   à  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ü ÜVVÌÌ)B*B4¸4¸>?I¢I¢ST^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÜÜVVÐÐ J J Ä Ä>>¸¸22¬¬&&   ” ”"#%„%„'ú'ú)p*p,æ,æ.\/\1Ò1Ò3H4H6¾6¾8494;ª;ª= > @–@–B C E‚E‚GøGøIPJPLÄLÄN8O8Q¬Q¬S T V²V²X&Y&[š[š]^`‚`‚böbödjejgÞgÞiRjRlÆlÆnDoDq¸q¸s,t,v v xy{ˆ{ˆ}ü}üp€p‚ä‚ä„X…X‡̇̉@Š@Œ´Œ´Ž(�(‘œ‘œ“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1Ÿ¾!1dŸ¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1Íç!1@Íç!1k("1¼k("11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1ŸQ"1˜ŸQ"1=’"1=’"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1…Ú"1Ì …Ú"1åþ"1H åþ"1',#1h',#1»7#1D»7#1e{#1Àe{#1Ž#1äŽ#1™#1À™#1“$1L“$1kò$1kò$1¥Þ%1¥Þ%1ì%1ôì%1É€&1€É€&1™R'1Ä™R'1Ae'1È€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ä_+ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 NŽ•ÕÉ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ä_+ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 NŽ•ÕÉ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ä_+  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 NŽ•ÕÉ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �½e…jŠ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 I‰Òl¬ bÒ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1<IÉ’Û[¤$nî7·€ÉI’Ò[6¶ � è h Á A þ> bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>-?-I¢I¢ST^Ž^ŽhisZsZ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbçbçd\e\gÑgÑiFjFl»l»n2o2q§q§stv‘v‘xy{{{{}ð}ðe€e‚Ú‚Ú„O…O‡¦‡¦‰ŠŒ�Œ�Ž�‘z‘z“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1_}!1è_}!1ý½!1dý½!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!11ç!1@1ç!1Ï'"1¼Ï'"1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1Q"1˜Q"1¡‘"1¡‘"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"1‹Ú"1Ì ‹Ú"1ëþ"1H ëþ"1-,#1h-,#1:#1D:#1Ãn#1ÐÃn#1A�#1ôA�#1?˜#1 ?˜#1¥$1¬¥$1}ò$1p}ò$1·Þ%1t·Þ%11ì%1T1ì%1{€&1à{€&1KR'1$KR'1ód'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;oÀŒ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 7w;{Ç bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;oÀŒ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 7w;{Ç bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1;oÀŒ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 7w;{Ç  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 °Ð«Ër’ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1;{¶ö1q¬ì µõ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d;»öv±1lì'§âbžYÙ”ÏOŠ EÅ ‚ v ö ½ = = ½ y ù  „ 4 ´ Ë K ï / j ª å % `   'g bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚRRÆÆ):*:4®4®>?I”I”ST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""ššŠŠ z z"ò"ò$f%f'Ú'Ú)N*N,Â,Â.6/61ª1ª346’6’89;z;z=î=î?b@bBÖBÖDJEJG¾G¾IJL…L…N÷N÷PiQiSÛSÛUlVlXÞXÞZP[P]Â]Â_4`4b¦b¦degŠgŠiüiüknlnnènèpZqZsÌsÌu>v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1Ÿ¾!1dŸ¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1Óç!1@Óç!1q("1¼q("11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1¥Q"1˜¥Q"1C’"1C’"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1…Ú"1Ì …Ú"1åþ"1H åþ"1',#1h',#1U:#1DU:#1ÿ}#1Àÿ}#1µŽ#1䵎#1³™#1À³™#1�$1L�$1Yò$1Yò$1“Þ%1“Þ%1 ì%1ô ì%1c€&1€c€&13R'1Ä3R'1Ûd'1È€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘Åa- b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O�•ÕÊ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘Åa- b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O�•ÕÊ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘Åa-  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O�•ÕÊ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ­ÍHhjŠ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 ;{¶úE  bÒ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1<;»öv±1lì'§âb�XØSŽÎš  s ó L Ì %¥—× bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÃÃ)8*84­4­>"?"I•I•ST^^hòhòrFsF} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD?I�I�ST^u^uhèhèr=s=} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……úúooääY Y"Î"Î$C%C'¸'¸)-*-,¢,¢./1Œ1Œ346v6v8ë8ë:`;`=Õ=Õ?H@HB»B»D.E.G¡G¡IJL‡L‡NúNúPmQmSàSàUSVSXÆXÆZ9[9]¬]¬_`b’b’degxgxiëiëk^l^nÑnÑpDqDs·s·u*v*x�x�z{}ƒ}ƒöö�i‚i„܄܆1‡1‰£‰£‹Œއއ�ù�ù’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1õ}!1Ðõ}!1“¾!1L“¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Çç!1(Çç!1e("1¤e("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1¯7#1,¯7#1é�#1˜é�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹T  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™ØÊ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹T  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™ØÊ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹T   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™ØÊ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 «Ëž¾lŒ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1[›€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1åe¼<“ j ê Ë  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á áWWÍÍ)C*C4¹4¹>-?-I¡I¡ST^‰^‰hýhýrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±''�� ‰ ‰"ÿ"ÿ$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@…@…BùBùDmEmGáGáIUJULÉLÉN=O=Q±Q±S%T%V™V™X Y [�[�]õ]õ_i`ibÝbÝdQeQgÅgÅi9j9l­l­n!o!q•q•s t v}v}xñxñze{e}Ù}ÙM€M‚Á‚Á„5…5‡©‡©‰ŠŒ�Œ�Ž�‘u‘u“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1e}!1èe}!1¾!1d¾!1kÅ!1lkÅ!1õÎ!1èõÎ!1+Ò!1´+Ò!17ç!1@7ç!1Õ'"1¼Õ'"1=/"1Ä=/"1Ç8"1@Ç8"1ý;"1 ý;"1 Q"1˜ Q"1§‘"1§‘"1™"1 ™"1™¢"1˜ ™¢"1Ï¥"1d Ï¥"1Ó"1ð Ó"1‘Ú"1Ì ‘Ú"1ñþ"1H ñþ"13,#1h3,#1:#1D:#1±}#1À±}#1e�#1äe�#1c˜#1Àc˜#1i$1Li$1Aò$1Aò$1{Þ%1{Þ%1õë%1ôõë%1ë&1€ë&1»Q'1Ä»Q'1cd'1È€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<pÁ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U;{Ò bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<pÁ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U;{Ò bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<pÁ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U;{Ò  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ³Óõt” bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7M7operation.4exchange :z2breasonOperatorInputsvalue8invocations1 R’ä(²ò  bÒ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r0_p64_c0_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M÷operation.5exchange :z2breasonOperatorInputsvalue9invocations1<RÒ$¤övÈHšlì>¾�â"t´ƒ  L Ì  • Þ ^ B‚ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×JJ½½)0*04£4£>?I…I…SöSö]g^ghØhØr*s*} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡úúmmààS S"Æ"Æ$9%9'¬'¬)*,’,’./1x1x3ë3ë5^6^8Ñ8Ñ:D;D=·=·?(@(B™B™D E G{G{IìIìK]L]NÎNÎP?Q?S°S°U!V!X’X’Z[]t]t_å_åaVbVdÇdÇf8g8i©i©kln‹n‹püpürmsmuÞuÞwOxOzÀzÀ|1}1¢¢�‚„„„„†Ö†ÖˆF‰F‹¶‹¶�&Ž&�–�–’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1û}!1Ðû}!1™¾!1L™¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Çç!1(Çç!1e("1¤e("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1I:#1,I:#1…Ž#1˜…Ž#1™$1$™$1qò$1èqò$1«Þ%1ì«Þ%1W€&1ÌW€&1'R'1'R'1Ïd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºV" b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ZšÚË  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºV" b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ZšÚË  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºV"  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ZšÚË   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 «Ë¦Æ\| bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ò2 €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1A Á  ˜ ï o Æ F b¢ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝQQÅÅ)9*94­4­>?I‘I‘ST^u^uhçhçrYsY} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��……ùùm m"á"á$U%U'É'É)=*=,±,±.%/%1™1™3 4 6�6�8õ8õ:i;i=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b‹b‹dýdýfogoiáiákSlSnÅnÅp7q7s©s©uvx�x�zÿzÿ|q}qãã�U‚U„DŽdž9‡9‰ª‰ª‹ŒŽŒŽŒ�ý�ý’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1k}!1Ðk}!1 ¾!1L ¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1=ç!1(=ç!1Û'"1¤Û'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1Q"1€Q"1­‘"1ü­‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#17#1,7#1ùŒ#1˜ùŒ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÄ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 -m=}à  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÄ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 -m=}à  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÄ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 -m=}à   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Œ¬=]Hh bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1$d€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1fæ1±ü|ÇG Ô  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚOOÄÄ)9*94®4®>$?$I˜I˜S T ^€^€hôhôrjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""šš„„ùùn n"ã"ã$X%X'Í'Í)B*B,·,·.,/,1¡1¡346‹6‹89;u;u=ê=ê?_@_BÔBÔDIEIG¾G¾I4J4L¨L¨NOQ�Q�STVxVxXìXìZ`[`]Ô]Ô_H`Hb¼b¼d0e0g¤g¤ijlŒlŒnoqtqtsèsèu\v\xÐxÐzD{D}¸}¸,€,‚ ‚ „…‡ЇЉÿ‰ÿ‹tŒtŽéŽé�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1ï}!1Ðï}!1�¾!1L�¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Áç!1(Áç!1_("1¤_("1[/"1¬[/"1å8"1(å8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1£7#1,£7#1¿�#1˜¿�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1#�&1Ì#�&1óR'1óR'1›e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®I b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 d¤•Õî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®I b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 d¤•Õî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®I  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 d¤•Õî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 «Ë—·^~ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1JŠ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÈH�ròG Ç šÚ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg à àYYÒÒ)K*K4Ä4Ä>=?=I´I´S)T)^œ^œhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggààYYÒÒ K K Ä Ä==¶¶//¨¨!!šš Œ Œ"#%~%~'÷'÷)p*p,é,é.b/b1Û1Û3T4T6Í6Í8F9F;¿;¿=8>8@±@±B*C*E£E£GHJ•J•L M OƒOƒQúQúSqTqVèVèX_Y_[Ö[Ö]M^M`Ä`Äb9c9e®e®g#h#j˜j˜l m o€o€qóqósftfvÙvÙxLyL{¿{¿}2~2€¥€¥‚ƒ…‹…‹‡þ‡þ‰pŠpŒâŒâŽT�T‘ƑƓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1k}!1Ðk}!1 ¾!1L ¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!17ç!1(7ç!1Õ'"1¤Õ'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1 Q"1€ Q"1§‘"1ü§‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1:#1,:#1ƒ�#1˜ƒ�#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1u€&1Ìu€&1ER'1ER'1íd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÂŽ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ,l<|Þ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÂŽ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ,l<|Þ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÂŽ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ,l<|Þ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ®Î:lŒ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¯ï €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1’  ] Ý ( ¨ ó s _Ÿ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØMMÂÂ)7*74¬4¬>!?!I”I”ST^x^xhêhêr^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?M@MBÂBÂD7E7G¬G¬I!J!L”L”NOQzQzSíSíU`V`XÓXÓZF[F]¹]¹_,`,bžbždeg‚g‚iôiôkflfnØnØpJqJs¼s¼u.v.x x z{}„}„öö�h‚h„Ú„Ú†N‡N‰Á‰Á‹4Œ4ާާ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1õ}!1Ðõ}!1“¾!1L“¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Áç!1(Áç!1_("1¤_("1[/"1¬[/"1å8"1(å8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#17:#1,7:#1UŽ#1˜UŽ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1±€&1̱€&1�R'1�R'1)e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯K b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 e¥•Õð0 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯K b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 e¥•Õð0 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯K  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 e¥•Õð0  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 «Ë›»Ff bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ß €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 œ ñ q Æ F ›  /o €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>/?/I¤I¤ST^ˆ^ˆhùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbçbçdZeZgÍgÍi@j@l³l³n$o$q•q•stvwvwxèxèzY{Y}Ê}Ê;€;‚¬‚¬„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1q}!1Ðq}!1¾!1L¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1Cç!1(Cç!1á'"1¤á'"1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1Q"1€Q"1³‘"1ü³‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1y7#1,y7#1�#1˜�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1ç€&1Ìç€&1·R'1·R'1_e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÅ‘ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IS“@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÅ‘ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IS“@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÅ‘  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IS“@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ƒ£6VRr bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1_€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1mí:º‡ÔT ï / €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÇÇ)>*>4µ4µ>.?.I¥I¥ST^Ž^Žhisrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝIVJVLÍLÍNDODQ»Q»S2T2V©V©X Y [—[—]^`…`…bübüdsesgêgêiajalØlØnJoJq¼q¼s.t.v v xy{„{„}ö}öh€h‚Ú‚Ú„L…L‡¾‡¾‰0Š0Œ¢Œ¢Ž�‘†‘†“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1é}!1Ðé}!1‡¾!1L‡¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1»ç!1(»ç!1Y("1¤Y("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1�Q"1€�Q"1+’"1ü+’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1—7#1,—7#1��#1˜��#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%15�&1Ì5�&1S'1S'1­e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£>  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 d¤¾þâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£>  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 d¤¾þâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£>   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 d¤¾þâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 «Ë�°Pp bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations19y€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1«+~þQÑ$ ¤ i© €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßWWÏÏ)G*G4¿4¿>7?7I­I­S#T#^™^™his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ��  "÷"÷$o%o'ç'ç)_*_,×,×.O/O1Ç1Ç3?4?6·6·8/9/;§;§=>@—@—BCE‡E‡GÿGÿIwJwLíLíNcOcQÙQÙSOTOVÅVÅX;Y;[±[±]'^'`�`�bce‰e‰gÿgÿiujulëlënaoaq×q×sMtMvÃvÃx9y9{¯{¯}%~%€›€›‚ƒ…‡…‡‡ý‡ý‰sŠsŒéŒéŽ_�_‘Õ‘Õ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1q}!1Ðq}!1¾!1L¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1=ç!1(=ç!1Û'"1¤Û'"1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1Q"1€Q"1­‘"1ü­‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1:#1,:#1§�#1˜§�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÃ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JR’þ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÃ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JR’þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÃ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JR’þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¡Á=]Rr bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¬ì €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¡ ! n î ; »  ˆ |¼ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb × ×LLÁÁ)6*64«4«>"?"I—I—S T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbb××LLÁÁ 6 6 « «  ••  ôôiiÞÞS S"È"È$=%='²'²)'*',œ,œ./1†1†3û3û5p6p8å8å:Z;Z=Ï=Ï?D@DB¹B¹D.E.G£G£IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1é}!1Ðé}!1‡¾!1L‡¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1»ç!1(»ç!1Y("1¤Y("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1�Q"1€�Q"1+’"1ü+’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1+:#1,+:#1%Ž#1˜%Ž#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤@  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 e¥¿ÿä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤@  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 e¥¿ÿä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤@   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 e¥¿ÿä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 «Ë�°0P bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ì €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1÷ w Ê J �  p ð ü< €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÇÇ)=*=4³4³>)?)I�I�ST^…^…hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒuŒuŽéŽé�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1w}!1Ðw}!1¾!1L¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1Cç!1(Cç!1á'"1¤á'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1Q"1€Q"1³‘"1ü³‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1…7#1,…7#1q�#1˜q�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1ù€&1Ìù€&1ÉR'1ÉR'1qe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÆ’ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 "b;{É  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÆ’ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 "b;{É  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÆ’  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 "b;{É   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 zš$DKk bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1"b€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ˆ[Û.® � B‚ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff à àZZÐÐ)F*F4¼4¼>2?2I¦I¦ST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffààZZÔÔ N N È ÈBB¼¼66°°**¤¤ ˜ ˜"#%ˆ%ˆ'þ'þ)t*t,ê,ê.`/`1Ö1Ö3L4L6Â6Â8898;®;®=$>$@š@šBCE†E†GüGüIrJrLæLæNZOZQÎQÎSBTBV¶V¶X*Y*[ž[ž]^`†`†búbúdnengâgâiVjVlÊlÊn>o>q²q²s&t&všvšxy{‚{‚}ö}öj€j‚Þ‚Þ„R…R‡ƇƉ:Š:Œ®Œ®Ž"�"‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1ã}!1Ðã}!1�¾!1L�¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1µç!1(µç!1S("1¤S("1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1‡Q"1€‡Q"1%’"1ü%’"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1‹7#1,‹7#1e�#1˜e�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ÿ€&1Ìÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜3ÿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M�ș٠bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜3ÿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M�ș٠bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜3ÿ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M�ș٠ bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 «Ë‰©>^ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1(h€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ž_ß0° � 8x €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Þ ÞUUÌÌ)C*C4º4º>1?1I¦I¦ST^�^�hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIYJYLÎLÎNCOCQ¸Q¸S-T-V¢V¢XY[Œ[Œ]^`v`vbëbëd`e`gÕgÕiJjJl¿l¿n4o4q©q©stv“v“xy{}{}}ò}òg€g‚܂܄Q…Q‡ƇƉ;Š;Œ°Œ°Ž%�%‘š‘š“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1q}!1Ðq}!1¾!1L¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1Cç!1(Cç!1á'"1¤á'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1Q"1€Q"1³‘"1ü³‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1mÚ"1´ mÚ"1Íþ"10 Íþ"1,#1P,#1:#1,:#1Ž#1˜Ž#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÄ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 !a;{Ê  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÄ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 !a;{Ê  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÄ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 !a;{Ê   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ”´Bb8X bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1µõ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ô T § ' z ú M Í Õ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÊÊ)>*>4²4²>&?&I˜I˜S T ^|^|hîhîr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$j%j'Þ'Þ)R*R,Æ,Æ.:/:1®1®3"4"6–6–8 9 ;~;~=ò=ò?f@fBÚBÚDNENGÂGÂI6J6L¨L¨NOQŒQŒSþSþUpVpXâXâZT[T]Æ]Æ_8`8bªbªdegŽgŽijlrlrnänäpVqVsÈsÈu:v:x¬x¬z{}�}�€‚t‚t„æ„æ†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wºoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z9z="=zA"AzE*’ªreasonOperatorKernelvalue10kernelSpecoutputElements576input0Elements640input1Elements62722Planned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })binvocations1·l7i!1ã}!1Ðã}!1�¾!1L�¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1µç!1(µç!1S("1¤S("1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1‡Q"1€‡Q"1%’"1ü%’"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1CÚ"1´ CÚ"1£þ"10 £þ"1å+#1På+#1:#1,:#1õ�#1˜õ�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™5 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 LŒÇšÚ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™5 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 LŒÇšÚ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™5  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 LŒÇšÚ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 «Ë…¥: bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¹ù €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ò R £ # t ô E Å É  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜRRÇÇ)<*<4±4±>&?&I™I™S T ^^hòhòrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒxxîîd d"Ú"Ú$O%O'Ä'Ä)9*9,®,®.#/#1˜1˜3 4 6‚6‚8÷8÷:l;l=á=á?V@VBËBËD@E@GµGµI*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b¨b¨degŽgŽijltltnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1}}!1Ð}}!1¾!1L¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Iç!1(Iç!1ç'"1¤ç'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1Q"1€Q"1¹‘"1ü¹‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1=Ú"1´ =Ú"1�þ"10 �þ"1ß+#1Pß+#1�7#1,�7#17Ž#1˜7Ž#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÇ“ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @<|Í  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÇ“ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @<|Í  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÇ“  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @<|Í   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 q‘­Í2 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations18x€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1à`¾>œ z ú H €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84­4­> ? I“I“ST^y^yhìhìr_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?L@LB¿B¿D2E2G¥G¥IJL‹L‹NþNþPqQqSäSäUWVWXÊXÊZ=[=]°]°_#`#b–b–d e g|g|iïiïkblbnÕnÕpHqHs»s»u.v.x¡x¡z{}‡}‡úú�m‚m„à„à†S‡S‰ʼnŋ7Œ7ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1Ý}!1ÐÝ}!1{¾!1L{¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1©ç!1(©ç!1G("1¤G("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1{Q"1€{Q"1’"1ü’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1—7#1,—7#1Ñ�#1˜Ñ�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1á€&1Ìá€&1±R'1±R'1Ye'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�(ô b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 P�Ë µõ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�(ô b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 P�Ë µõ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�(ô  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 P�Ë µõ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‚¢,LXx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations10p€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1À@–lìB   à €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk å å[[ÑÑ)G*G4½4½>1?1I¥I¥ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkåå__ÙÙ S S Í ÍGGÁÁ;;µµ++¡¡ � �"#%y%y'ï'ï)e*e,Û,Û.Q/Q1Ç1Ç3=4=6³6³8)9);Ÿ;Ÿ=>@‰@‰BýBýDqEqGåGåIYJYLÍLÍNAOAQµQµS)T)V�V�XY[…[…]ù]ù_m`mbábádUeUgÉgÉi=j=l±l±n%o%q™q™s t v�v�xõxõzi{i}Ý}ÝQ€Q‚łń9…9‡­‡­‰ Š Œ“Œ“Ž�‘y‘y“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1w}!1Ðw}!1¾!1L¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Iç!1(Iç!1ç'"1¤ç'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1Q"1€Q"1¹‘"1ü¹‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#1=:#1,=:#1ÙŽ#1˜ÙŽ#1Ÿ$1$Ÿ$1wò$1èwò$1±Þ%1ì±Þ%1!€&1Ì!€&1ñQ'1ñQ'1™d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÅ‘ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @<|Î bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÅ‘ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @<|Î bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÅ‘  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @<|Î  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ˆ¨«Ë? bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ö €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1X Ø 6 ¶  ” ò r ¦æ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛNNÁÁ)4*44§4§>?I‰I‰SúSú]k^khÜhÜrMsM} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹þþqqääW W"Ê"Ê$=%='°'°)#*#,–,–. / 1|1|3ï3ï5b6b8Õ8Õ:H;H=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éaZbZdËdËf#?#I•I•ST^y^yhëhër]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‰‰ýýq q"å"å$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=á=á?S@SBÅBÅD7E7G©G©IJL�L�NÿNÿPqQqSãSãUUVUXÇXÇZ9[9]«]«_`b�b�degsgsiåiåkWlWnÉnÉp;q;s­s­uvx‘x‘z{}u}uçç�Y‚Y„˄ˆ=‡=‰®‰®‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1}}!1Ð}}!1¾!1L¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Oç!1(Oç!1í'"1¤í'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1!Q"1€!Q"1¿‘"1ü¿‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"11Ú"1´ 1Ú"1‘þ"10 ‘þ"1Ó+#1PÓ+#1µ7#1,µ7#1�#1˜�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÈ” b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W;{Ï bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÈ” b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W;{Ï bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÈ”  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W;{Ï  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 «Ëhˆ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1NŽ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18¸!¡ Š ó s Î €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞSSÈÈ)=*=4²4²>(?(IœIœST^„^„høhørnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžˆˆýýr r"ç"ç$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡އމŠŒxŒxŽíŽí�b‘b“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1×}!1Ð×}!1u¾!1Lu¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1£ç!1(£ç!1A("1¤A("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1uQ"1€uQ"1’"1ü’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#1¯7#1,¯7#1�Ž#1˜�Ž#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%15�&1Ì5�&1S'1S'1­e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚é b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“Τä bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚é b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“Τä bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚é  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“Î¤ä  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 {›:Qq bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1F†€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1˜ùyÚZ » ; f¦ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk ä ä]]ÖÖ)O*O4È4È>A?AI¸I¸S-T-^ ^ his†s†} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkää]]ÖÖ O O È ÈAAºº33¬¬%%žž � �" # %‚%‚'û'û)t*t,í,í.f/f1ß1ß3X4X6Ñ6Ñ8J9J;Ã;Ã=<><@µ@µB.C.E§E§G H J™J™LMO‡O‡QþQþSuTuVìVìXcYc[Ú[Ú]Q^Q`È`Èb=c=e²e²g'h'jœjœlmo„o„q÷q÷sjtjvÝvÝxPyP{Ã{Ã}6~6€©€©‚ƒ…�…�‡ˆŠtŠtŒæŒæŽX�X‘ʑʓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1}}!1Ð}}!1¾!1L¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Oç!1(Oç!1í'"1¤í'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1!Q"1€!Q"1¿‘"1ü¿‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"17Ú"1´ 7Ú"1—þ"10 —þ"1Ù+#1PÙ+#1a:#1,a:#1«�#1˜«�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÆ’ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V;{Ð bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÆ’ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V;{Ð bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÆ’  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V;{Ð  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 «Ëzš$ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1÷7 €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ü \ Å E ® . — w· €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜQQÆÆ);*;4°4°>%?%I˜I˜S T ^|^|hîhîrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒvvëë` `"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1×}!1Ð×}!1u¾!1Lu¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1©ç!1(©ç!1G("1¤G("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1{Q"1€{Q"1’"1ü’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1O:#1,O:#19�#1˜9�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒë b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R’Í ¥å bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒë b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R’Í ¥å bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒë  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R’Í ¥å  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 o�îXx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ç' €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1œ  } ý ^ Þ ? ¿ G €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii à àWWÎÎ)E*E4¼4¼>3?3I¨I¨ST^Œ^Œhýhýroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiààWWÎÎ E E ¼ ¼33ªª!!˜˜††ýýt t"ë"ë$b%b'Ù'Ù)P*P,Ç,Ç.>/>1µ1µ3,4,6£6£89;‘;‘=>@@BöBöDmEmGäGäI[J[LÐLÐNEOEQºQºS/T/V¤V¤XY[Ž[Ž]^`x`xbëbëd^e^gÑgÑiDjDl·l·n(o(q™q™s t v{v{xìxìz]{]}Î}Î?€?‚°‚°„!…!‡“‡“‰ŠŒuŒu޿޿�W‘W“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ý}!1�Ý}!1{¾!1{¾!1­Â!1$­Â!1©ç!1�©ç!1G("1G("1,"1$,"1{Q"1�{Q"1’"1’"1Q–"1$Q–"1çØ"1�çØ"1Gý"1üGý"1)0#1 )0#1õ~#1| õ~#1á$1Ü á$1¹ò$1  ¹ò$1óÞ%1¤ óÞ%1�&1„ �&1áR'1ÈáR'1‰e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19mS‡Ú¦!í b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19mS‡Ú¦!í b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19mS‡Ú¦!í b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«Ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww î îeeÜÜ)S*S4Ê4Ê>A?AI¸I¸S/T/^¡^¡his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwîîeeÜÜ S S Ê ÊAA¸¸//¦¦””  ‚ ‚"ù"ù$p%p'ç'ç)^*^,Õ,Õ.L/L1Ã1Ã3:4:6±6±8(9(;Ÿ;Ÿ=>@�@�BCE{E{GòGòIiJiLàLàNWOWQÎQÎSETEV¼V¼X3Y3[ª[ª]!^!`˜`˜bce†e†gýgýitjtlëlën]o]qÏqÏsAtAv³v³x%y%{—{—} ~ €{€{‚í‚í„_…_‡чщCŠCŒµŒµŽ'�'‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ñ}!1�Ñ}!1o¾!1 o¾!1qÅ!1qÅ!1£ç!1€£ç!1A("1üA("1C/"1C/"1uQ"1puQ"1’"1ì’"1™"1ô™"1…Ú"1`…Ú"1åþ"1Ìåþ"1)0#1ì)0#1õ~#1L õ~#1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1S�&1T S�&1#S'1˜#S'1Ëe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÝ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Á€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÝ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Á€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÝ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Á €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Gg  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ì ìddÜÜ)T*T4Ì4Ì>B?BI¸I¸S.T.^¤^¤his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttììddÜÜ T T Ì ÌDD¼¼44¬¬$$œœ Œ Œ"#%|%|'ô'ô)l*l,ä,ä.\/\1Ô1Ô3L4L6Ä6Ä8<9<;´;´=,>,@¤@¤BCE”E”G H J‚J‚LøLøNnOnQäQäSZTZVÐVÐXFYF[¼[¼]2^2`¨`¨bce”e”g h j€j€lölönlolqâqâsXtXvÎvÎxDyD{º{º}0~0€¦€¦‚ƒ…’…’‡ˆŠ~Š~ŒôŒôŽj�j‘à‘à“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ý}!1�Ý}!1{¾!1{¾!1­Â!1$­Â!1¯ç!1�¯ç!1M("1M("1,"1$,"1�Q"1��Q"1’"1’"1Q–"1$Q–"1çØ"1�çØ"1Gý"1üGý"1)0#1 )0#1õ~#1| õ~#1Ï$1Ü Ï$1§ò$1  §ò$1áÞ%1¤ áÞ%1·€&1„ ·€&1‡R'1ȇR'1/e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:nQ…Ö¢$ð b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:nQ…Ö¢$ð b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:nQ…Ö¢$ð b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«Ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu ê ê__ÔÔ)I*I4¾4¾>3?3I¨I¨ST^�^�hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuêê__ÔÔ I I ¾ ¾33¨¨’’||ññf f"Û"Û$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I+J+L L NOQŠQŠSÿSÿUtVtXéXéZ^[^]Ó]Ó_H`Hb½b½d2e2g§g§ijl‘l‘noququsçsçuYvYxËxËz={=}¯}¯!€!‚“‚“„…‡w‡w‰é‰é‹[Œ[ŽÍŽÍ�?‘?“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ñ}!1�Ñ}!1o¾!1 o¾!1qÅ!1qÅ!1£ç!1€£ç!1A("1üA("1C/"1C/"1uQ"1puQ"1’"1ì’"1™"1ô™"1‹Ú"1`‹Ú"1ëþ"1Ìëþ"1)0#1ì)0#1õ~#1L õ~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1á€&1T á€&1±R'1˜±R'1Ye'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÝ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Á€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÝ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Á€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÝ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Á €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Pp  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ê ê``ÖÖ)L*L4Â4Â>6?6IªIªST^’^’hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttêê``ÖÖ L L  Â88®®$$šš††üür r"è"è$^%^'Ô'Ô)J*J,À,À.6/61¬1¬3"4"6˜6˜89;„;„=ú=ú?p@pBæBæD\E\GÒGÒIFJFLºLºN.O.Q¢Q¢STVŠVŠXþXþZr[r]æ]æ_Z`ZbÎbÎdBeBg¶g¶i*j*lžlžnoq†q†súsúunvnxâxâzV{V}Ê}Ê>€>‚²‚²„&…&‡š‡š‰ŠŒ‚Œ‚ŽöŽö�j‘j“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1•}!1�•}!13¾!1 3¾!1¿Â!1¿Â!1aç!1€aç!1ÿ'"1üÿ'"1‘,"1‘,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1c–"1ôc–"1çØ"1`çØ"1Gý"1ÌGý"1)0#1ì)0#1õ~#1L õ~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1�&1T �&1çR'1˜çR'1�e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„×£ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1W€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„×£ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1W€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„×£ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1W €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«Ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww ñ ñkkáá)W*W4Í4Í>A?AIµIµS)T)^�^�his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwññkkåå _ _ Ù ÙSSÍÍGGÁÁ;;µµ// © ©"###%™%™'(*…*…,û,û.q/q1ç1ç3]4]6Ó6Ó8I9I;¿;¿=5>5@«@«B!C!E—E—G H J�J�LõLõNiOiQÝQÝSQTQVÅVÅX9Y9[­[­]!^!`•`•b c e}e}gñgñiejelÙlÙnMoMqÁqÁs5t5v©v©xy{‘{‘}~€y€y‚í‚í„a…a‡Õ‡Õ‰IŠIŒ½Œ½Ž1�1‘¥‘¥“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1×}!1�×}!1u¾!1 u¾!1MÅ!1MÅ!1£ç!1€£ç!1A("1üA("1/"1/"1uQ"1puQ"1’"1ì’"1ñ˜"1ôñ˜"1yÚ"1`yÚ"1Ùþ"1ÌÙþ"1)0#1ì)0#1õ~#1L õ~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kæ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¤ä€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kæ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¤ä€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kæ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¤ä €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1@`  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp ç ç^^ÕÕ)L*L4Ã4Ã>8?8I­I­S"T"^—^—h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppçç^^ÕÕ L L à Ã::±±((ŸŸ�� { {"ò"ò$i%i'à'à)W*W,Î,Î.E/E1¼1¼33436ª6ª8!9!;˜;˜=>@†@†BýBýDtEtGëGëI`J`LÕLÕNJOJQ¿Q¿S4T4V©V©XY[“[“]^`}`}bòbòdgeggÜgÜiQjQlÆlÆn;o;q°q°s%t%všvšxy{„{„}ù}ùn€n‚ã‚ã„X…X‡͇͉BŠBŒ·Œ·Ž,�,‘¡‘¡“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1�}!1��}!1-¾!1 -¾!1¿Â!1¿Â!1aç!1€aç!1ÿ'"1üÿ'"1‘,"1‘,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1c–"1ôc–"1çØ"1`çØ"1Gý"1ÌGý"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ÓŸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1V€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ÓŸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1V€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ÓŸ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1V €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«Ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu í íeeÙÙ)M*M4Á4Á>3?3I¥I¥ST^‰^‰hûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuííeeÝÝ U U Í ÍEE½½55­­%%�� � �"#%y%y'í'í)a*a,Õ,Õ.I/I1½1½31416¥6¥89;�;�=>@u@uBéBéD]E]GÑGÑICJCLµLµN'O'Q™Q™S T V}V}XïXïZa[a]Ó]Ó_E`Eb·b·d)e)g›g›i j llnñnñpcqcsÕsÕuGvGx¹x¹z+{+}�}�€‚�‚�„ó„ó†e‡e‰׉׋IŒI޻޻�-‘-“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1×}!1�×}!1u¾!1 u¾!1MÅ!1MÅ!1£ç!1€£ç!1A("1üA("1/"1/"1uQ"1puQ"1’"1ì’"1ñ˜"1ôñ˜"1‹Ú"1`‹Ú"1ëþ"1Ìëþ"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~æ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¥å€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~æ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¥å€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~æ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¥å €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Qq  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ç ç]]ÒÒ)G*G4¼4¼>/?/I¢I¢ST^ˆ^ˆhûhûrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqçç]]ÓÓ I I ¿ ¿55««!!——  ƒƒùùo o"å"å$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ì=ì?a@aBÖBÖDKEKGÀGÀI3J3L¦L¦NOQŒQŒSÿSÿUrVrXåXåZX[X]Ë]Ë_>`>b±b±d$e$g—g—i j l}l}nðnðpcqcsÖsÖuIvIx¼x¼z/{/}¢}¢€‚ˆ‚ˆ„û„û†n‡n‰á‰á‹TŒTŽÇŽÇ�:‘:“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1‰}!1�‰}!1'¾!1 '¾!1­Â!1­Â!1[ç!1€[ç!1ù'"1üù'"1,"1,"1-Q"1p-Q"1Ë‘"1ìË‘"1Q–"1ôQ–"1sÚ"1`sÚ"1Óþ"1ÌÓþ"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Ô  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Ô  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Ô  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1=]  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æ[[ÐÐ)E*E4¸4¸>+?+IžIžST^„^„h÷h÷rhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  €€õõj j"ß"ß$T%T'É'É)>*>,³,³.(/(1�1�346‡6‡8ü8ü:q;q=ä=ä?W@WBÊBÊD=E=G°G°I#J#L–L–N O Q|Q|SïSïUbVbXÕXÕZH[H]»]»_.`.b¡b¡deg‡g‡iúiúkmlmnànàpSqSsÆsÆu9v9x¬x¬z{}’}’€‚x‚x„ë„ë†\‡\‰Ή΋@Œ@޲޲�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Ý}!1�Ý}!1{¾!1 {¾!1qÅ!1qÅ!1©ç!1€©ç!1G("1üG("1C/"1C/"1{Q"1p{Q"1’"1ì’"1™"1ô™"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1ç€&1T ç€&1·R'1˜·R'1_e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆ#ï b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1É €} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆ#ï b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1É €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆ#ï b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1É  €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«Ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp ê ê``ÖÖ)L*L4À4À>4?4I¨I¨ST^�^�hisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppêêddÞÞ X X Ò ÒLLÆÆ@@ºº00¦¦ ’ ’"#%~%~'ô'ô)j*j,à,à.V/V1Ì1Ì3B4B6¸6¸8.9.;¤;¤=>@Œ@ŒBCEtEtGèGèI\J\LÐLÐNDODQ¸Q¸S,T,V V XY[ˆ[ˆ]ü]ü_p`pbäbädXeXgÌgÌi@j@l´l´n(o(qœqœstv„v„xøxøzl{l}à}àT€T‚ȂȄ<…<‡®‡®‰!Š!Œ”Œ”Ž�‘z‘z“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1‰}!1�‰}!1'¾!1 '¾!1³Â!1³Â!1Uç!1€Uç!1ó'"1üó'"1…,"1…,"1'Q"1p'Q"1Å‘"1ìÅ‘"1W–"1ôW–"1yÚ"1`yÚ"1Ùþ"1ÌÙþ"1#0#1ì#0#1ï~#1L ï~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1-€&1T -€&1ýQ'1˜ýQ'1¥d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KМ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KМ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KМ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Dd  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æYYÌÌ)?*?4°4°>!?!I’I’ST^t^thåhårTsT} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  ||ïïb b"Õ"Õ$H%H'»'»).*.,¡,¡./1‡1‡3ú3ú5m6m8à8à:S;S=Ä=Ä?5@5B¦B¦DEGˆGˆIùIùKjLjNÛNÛPLQLS½S½U.V.XŸXŸZ[]�]�_ò_òacbcdÔdÔfEgEi¶i¶k'l'n˜n˜p q szszuëuëw\x\zÍzÍ|>}>¯¯� ‚ „‘„‘†‡‰p‰p‹à‹à�PŽP�À�À’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Ý}!1�Ý}!1{¾!1 {¾!1qÅ!1qÅ!1©ç!1€©ç!1G("1üG("1C/"1C/"1{Q"1p{Q"1’"1ì’"1™"1ô™"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1o€&1T o€&1?R'1˜?R'1çd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡#ï b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1È€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡#ï b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1È€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡#ï b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1È €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«Ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp è è\\ÐÐ)D*D4¶4¶>(?(IšIšS T ^~^~hðhðr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppèè``ØØ P P È È@@¸¸00¨¨�� x x"ì"ì$`%`'Ô'Ô)H*H,¼,¼.0/01¤1¤346Œ6Œ89;t;t=æ=æ?X@XBÊBÊD[>]°]°_"`"b”b”degxgxiêiêk\l\nÎnÎp@q@s²s²u$v$x–x–z{}z}zìì�^‚^„ЄІ@‡@‰±‰±‹"Œ"ޓޓ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ƒ}!1�ƒ}!1!¾!1 !¾!1ÑÂ!1ÑÂ!1Uç!1€Uç!1ó'"1üó'"1£,"1£,"1'Q"1p'Q"1Å‘"1ìÅ‘"1u–"1ôu–"1sÚ"1`sÚ"1Óþ"1ÌÓþ"10#1ì0#1é~#1L é~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1Ï€&1T Ï€&1ŸR'1˜ŸR'1Ge'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ñ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1.n€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ñ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1.n€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ñ� b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1.n €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Bb  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo ç ç\\ÑÑ)F*F4»4»>/?/I£I£ST^‹^‹hÿhÿrsss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooçç__×× O O Ç Ç??··//§§‘‘ { {"ð"ð$e%e'Ú'Ú)O*O,Ä,Ä.9/91®1®3#4#6˜6˜8 9 ;‚;‚=÷=÷?l@lBáBáDVEVGËGËI?J?L³L³N'O'Q›Q›STVƒVƒX÷X÷Zk[k]ß]ß_S`SbÇbÇd;e;g¯g¯i#j#l—l—n o qqsósóugvgxÛxÛzO{O}Ã}Ã7€7‚«‚«„…‡“‡“‰ŠŒ}Œ}ŽòŽò�g‘g“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1Ý}!1�Ý}!1{¾!1 {¾!15Å!15Å!1¯ç!1€¯ç!1M("1üM("1/"1/"1�Q"1p�Q"1’"1ì’"1Ù˜"1ôÙ˜"1ÛØ"1`ÛØ"1;ý"1Ì;ý"10#1ì0#1é~#1L é~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1/�&1T /�&1ÿR'1˜ÿR'1§e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘,ø b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1•Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘,ø b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1•Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘,ø b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1•Õ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«Ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll å å^^××)P*P4É4É>@?@I·I·S,T,^Ÿ^Ÿhisƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllåå^^×× P P É ÉBB»»44­­&&ŸŸ ‘ ‘" # %ƒ%ƒ'ü'ü)u*u,î,î.g/g1à1à3Y4Y6Ò6Ò8K9K;Ä;Ä==>=@¶@¶B/C/E¨E¨G!H!J˜J˜LMO†O†QýQýStTtVëVëXbYb[Ù[Ù]P^P`Ç`Çb*?*I�I�ST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““}}òòg g"Ü"Ü$Q%Q'Æ'Æ);*;,°,°.%/%1š1š346„6„8ù8ù:n;n=ã=ã?X@XBÍBÍDBEBG·G·I*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b§b§deg‹g‹iýiýkolonánápSqSsÅsÅu7v7x©x©z{}�}�ÿÿ�q‚q„ã„ã†U‡U‰ȉȋ;Œ;ޮޮ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1Ý}!1�Ý}!1{¾!1 {¾!15Å!15Å!1¯ç!1€¯ç!1M("1üM("1/"1/"1�Q"1p�Q"1’"1ì’"1Ù˜"1ôÙ˜"1ÛØ"1`ÛØ"1;ý"1Ì;ý"10#1ì0#1é~#1L é~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1½€&1T ½€&1�R'1˜�R'15e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�,ø b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1•Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�,ø b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1•Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�,ø b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1•Õ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«Ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ã ãZZÑÑ)H*H4¿4¿>4?4I©I©ST^�^�hþhþrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllããZZÑÑ H H ¿ ¿66­­$$››‰‰ w w"î"î$e%e'Ü'Ü)S*S,Ê,Ê.A/A1¸1¸3/4/6¦6¦89;”;”= > @‚@‚BùBùDpEpGçGçI\J\LÑLÑNFOFQ»Q»S0T0V¥V¥XY[�[�]^`y`ybìbìd_e_gÒgÒiEjEl¸l¸n)o)qšqšs t v|v|xíxíz^{^}Ï}Ï@€@‚±‚±„"…"‡’‡’‰ŠŒtŒtŽåŽå�V‘V“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1}}!1�}}!1¾!1 ¾!1­Â!1­Â!1Oç!1€Oç!1í'"1üí'"1,"1,"1!Q"1p!Q"1¿‘"1ì¿‘"1Q–"1ôQ–"1gÚ"1`gÚ"1Çþ"1ÌÇþ"10#1ì0#1é~#1L é~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ó€&1T ó€&1ÃR'1˜ÃR'1ke'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Κ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Κ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Κ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1;[  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk â âYYÐÐ)G*G4¾4¾>5?5I¬I¬S#T#^•^•hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkââYYÐÐ G G ¾ ¾55¬¬##ššˆˆÿÿv v"í"í$d%d'Û'Û)R*R,É,É.@/@1·1·3.4.6¥6¥89;“;“= > @�@�BøBøDoEoGæGæI]J]LÔLÔNKOKQÂQÂS9T9V°V°X'Y'[ž[ž]^`Œ`ŒbcezezgñgñihjhlßlßnQoQqÃqÃs5t5v§v§xy{‹{‹}ý}ýo€o‚á‚á„S…S‡Ňʼn7Š7Œ©Œ©Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ã}!1�ã}!1�¾!1 �¾!1eÅ!1eÅ!1µç!1€µç!1S("1üS("17/"17/"1‡Q"1p‡Q"1%’"1ì%’"1 ™"1ô ™"1ÛØ"1`ÛØ"1;ý"1Ì;ý"10#1ì0#1é~#1L é~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1;�&1T ;�&1 S'1˜ S'1³e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš5 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Æ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš5 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Æ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš5 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Æ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«Ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àXXÐÐ)H*H4À4À>6?6I¬I¬S"T"^˜^˜his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  �� € €"ø"ø$p%p'è'è)`*`,Ø,Ø.P/P1È1È3@4@6¸6¸8090;¨;¨= > @˜@˜BCEˆEˆGHJvJvLìLìNbObQØQØSNTNVÄVÄX:Y:[°[°]&^&`œ`œbceˆeˆgþgþitjtlêlên`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1w}!1�w}!1¾!1 ¾!1­Â!1­Â!1Iç!1€Iç!1ç'"1üç'"1,"1,"1Q"1pQ"1¹‘"1칑"1Q–"1ôQ–"1…Ú"1`…Ú"1åþ"1Ìåþ"10#1ì0#1é~#1L é~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÊ– b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÊ– b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÊ– b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Xx  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii Þ ÞSSÈÈ)=*=4²4²>'?'IœIœST^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiÞÞSSÈÈ = = ² ²''œœ††ûûppååZ Z"Ï"Ï$D%D'¹'¹).*.,£,£./1�1�346w6w8ì8ì:a;a=Ö=Ö?K@KBÀBÀD5E5GªGªIJL”L”N O Q~Q~SóSóUhVhXÝXÝZR[R]Ç]Ç_<`*?*IžIžST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽzzððf f"Ü"Ü$R%R'È'È)>*>,´,´.*/*1 1 346Œ6Œ89;x;x=î=î?d@dBÚBÚDPEPGÆGÆI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1w}!1�w}!1¾!1 ¾!1ÅÂ!1ÅÂ!1Iç!1€Iç!1ç'"1üç'"1—,"1—,"1Q"1pQ"1¹‘"1칑"1i–"1ôi–"1aÚ"1`aÚ"1Áþ"1ÌÁþ"10#1ì0#1ã~#1L ã~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1ù€&1T ù€&1ÉR'1˜ÉR'1qe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxË— b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1+k€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxË— b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1+k€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxË— b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1+k €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations16V  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk å å__ÕÕ)K*K4Á4Á>5?5I©I©ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkåå__ÙÙ S S Í ÍGGÁÁ;;µµ//©©## � �"#%�%�'(*y*y,ï,ï.e/e1Û1Û3Q4Q6Ç6Ç8=9=;³;³=)>)@Ÿ@ŸBCE‹E‹GHJuJuLéLéN]O]QÑQÑSETEV¹V¹X-Y-[¡[¡]^`‰`‰býbýdqeqgågåiYjYlÍlÍnAoAqµqµs)t)v�v�xy{…{…}ù}ùm€m‚á‚á„U…U‡ɇɉ=Š=Œ±Œ±Ž%�%‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1é}!1�é}!1‡¾!1 ‡¾!1/Å!1/Å!1»ç!1€»ç!1Y("1üY("1/"1/"1�Q"1p�Q"1+’"1ì+’"1Ó˜"1ôÓ˜"1ÕØ"1`ÕØ"15ý"1Ì5ý"10#1ì0#1ã~#1L ã~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1ù€&1T ù€&1ÉR'1˜ÉR'1qe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£>  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1•Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£>  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1•Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£>  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1•Õ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«Ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>,?,I¡I¡ST^‹^‹hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbæbæd[e[gÐgÐiEjElºlºn/o/q¤q¤stvŽvŽxy{x{x}í}íb€b‚ׂׄL…L‡Á‡Á‰6Š6Œ«Œ«Ž � ‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1q}!1�q}!1¾!1 ¾!1ÅÂ!1ÅÂ!1Cç!1€Cç!1á'"1üá'"1—,"1—,"1Q"1pQ"1³‘"1쳑"1i–"1ôi–"1Ú"1`Ú"1ßþ"1Ìßþ"10#1ì0#1ã~#1L ã~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÇ“ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1,l€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÇ“ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1,l€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÇ“ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1,l €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ww  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áYYÍÍ)A*A4µ4µ>'?'I™I™S T ^}^}hïhïrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‘‘  � �"ù"ù$m%m'á'á)U*U,É,É.=/=1±1±3%4%6™6™8 9 ;�;�=õ=õ?i@iBÝBÝDQEQGÅGÅI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1é}!1�é}!1‡¾!1 ‡¾!1/Å!1/Å!1»ç!1€»ç!1Y("1üY("1/"1/"1�Q"1p�Q"1+’"1ì+’"1Ó˜"1ôÓ˜"1ÕØ"1`ÕØ"15ý"1Ì5ý"10#1ì0#1ã~#1L ã~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢>  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1•Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢>  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1•Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢>  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1•Õ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«Ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÆÆ);*;4°4°>#?#I–I–S T ^|^|hïhïrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnänäpWqWsÊsÊu=v=x°x°z#{#}–}– € ‚|‚|„ï„ï†b‡b‰Õ‰Õ‹HŒH޻޻�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1q}!1�q}!1¾!1 ¾!1¡Â!1¡Â!1=ç!1€=ç!1Û'"1üÛ'"1s,"1s,"1Q"1pQ"1­‘"1ì­‘"1E–"1ôE–"1ÏØ"1`ÏØ"1/ý"1Ì/ý"10#1ì0#1Ý~#1L Ý~#1¥$1¬ ¥$1}ò$1p }ò$1·Þ%1t ·Þ%1u€&1T u€&1ER'1˜ER'1íd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÈ” b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÈ” b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÈ” b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«Ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94¬4¬>?I’I’ST^z^zhíhír^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ø=Ø?K@KB¾B¾D1E1G¤G¤IJLŠLŠNýNýPpQpSãSãUVVVXÉXÉZ<[<]¯]¯_"`"b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†R‡R‰ĉċ6Œ6ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1ï}!1�ï}!1�¾!1 �¾!1YÅ!1YÅ!1Áç!1€Áç!1_("1ü_("1+/"1+/"1“Q"1p“Q"11’"1ì1’"1ý˜"1ôý˜"1Ú"1`Ú"1yþ"1Ìyþ"10#1ì0#1Ý~#1L Ý~#1½$1¬ ½$1åð$1p åð$1ÃÀ%1t ÃÀ%1Õ€&1T Õ€&1ËP'1˜ËP'1sc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬G b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1À} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬G b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1À} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬G b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1à €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Þ ÞTTÊÊ)@*@4´4´>(?(IœIœST^Ž^Žhistst} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÞÞXXÒÒ L L Æ Æ@@ºº44®®$$šš † †"ü"ü$r%r'è'è)^*^,Ô,Ô.J/J1À1À36466¬6¬8"9";˜;˜= > @€@€BôBôDhEhGÜGÜIPJPLÄLÄN8O8Q¬Q¬S T V”V”XY[|[|]ð]ð_d`dbâbâdVeVgÊgÊi>j>l²l²n&o&qšqšstv‚v‚xöxözj{j}Þ}ÞR€R‚ƂƄ:…:‡¬‡¬‰ŠŒ’Œ’Ž�‘x‘x“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1k}!1�k}!1 ¾!1  ¾!1¡Â!1¡Â!1=ç!1€=ç!1Û'"1üÛ'"1s,"1s,"1Q"1pQ"1­‘"1ì­‘"1E–"1ôE–"1ÏØ"1`ÏØ"1/ý"1Ì/ý"10#1ì0#1Ý~#1L Ý~#1“$1¬ “$1kò$1p kò$1¥Þ%1t ¥Þ%1€&1T €&1åQ'1˜åQ'1�d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÄ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÄ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÄ� b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«Ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚMMÀÀ)3*34¤4¤>?I†I†SûSû]l^lhÝhÝrLsL} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠýýppããV V"É"É$<%<'¯'¯)"*",•,•./1{1{3î3î5a6a8Ô8Ô:G;G=¸=¸?)@)BšBšD E G|G|IíIíK^L^NÏNÏP@Q@S±S±U"V"X“X“Z[]u]u_æ_æa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1ï}!1�ï}!1�¾!1 �¾!1YÅ!1YÅ!1Áç!1€Áç!1_("1ü_("1+/"1+/"1“Q"1p“Q"11’"1ì1’"1ý˜"1ôý˜"1=Ú"1`=Ú"1�þ"1Ì�þ"10#1ì0#1Ý~#1L Ý~#1¥$1¬ ¥$1Íð$1p Íð$1«À%1t «À%1]€&1T ]€&1SP'1˜SP'1ûb'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«G b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«G b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«G b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1=9  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜPPÄÄ)8*84ª4ª>?IŽIŽST^z^zhìhìr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœ„„øøl l"à"à$T%T'È'È)<*<,°,°.$/$1˜1˜3 4 6€6€8ô8ô:h;h=Ú=Ú?L@LB¾B¾D0E0G¢G¢IJL†L†NøNøPjQjSÜSÜUNVNXÀXÀZ2[2]¤]¤_`b�b�degtgtiæiækXlXnÊnÊprÅ‘ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1(h€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÅ‘ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1(h€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÅ‘ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1(h €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1·�'K  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛPPÇÇ)<*<4±4±>%?%I™I™ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››……úúo o"ä"ä$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`IbÉbÉd=e=g±g±i%j%l™l™n o q�q�sõsõuivixÝxÝzQ{Q}Å}Å9€9‚­‚­„!…!‡•‡•‰ Š ŒŒŽôŽô�i‘i“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1õ}!1�õ}!1“¾!1 “¾!1)Å!1)Å!1Áç!1€Áç!1_("1ü_("1û."1û."1“Q"1p“Q"11’"1ì1’"1͘"1ô͘"1«Ø"1`«Ø"1 ý"1Ì ý"10#1ì0#1Ý~#1L Ý~#1Ï$1¬ Ï$1÷ð$1p ÷ð$1ÕÀ%1t ÕÀ%1)�&1T )�&1Q'1˜Q'1Çc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µP b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1•Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µP b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1•Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µP b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1•Õ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«§  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ù ÙRRÕÕ)N*N4Ç4Ç>>?>IµIµS*T*^©^©his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÙÙRRËË D D ½ ½66¯¯((¡¡““  … …"þ"þ$�%�'ú'ú)s*s,ì,ì.e/e1Þ1Þ3W4W6Ð6Ð8I9I;Â;Â=;>;@´@´B-C-E¦E¦GHJ–J–L M O„O„QûQûSrTrVéVéX`Y`[×[×]N^N`Å`Åb:c:e¯e¯g$h$j™j™lmo�o�qrtstsvævæxYyY{Ì{Ì}?~?€²€²‚%ƒ%…˜…˜‡ ˆ Š{Š{ŒíŒíŽ_�_‘ёѓ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1e}!1�e}!1¾!1 ¾!1¹Â!1¹Â!17ç!1€7ç!1Õ'"1üÕ'"1‹,"1‹,"1 Q"1p Q"1§‘"1ì§‘"1]–"1ô]–"1sÚ"1`sÚ"1Óþ"1ÔÓþ"10#1ô0#1Ý~#1T Ý~#1«$1´ «$1Óð$1x Óð$1±À%1| ±À%1�€&1\ �€&1wP'1 wP'1c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<pÁ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1)i€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<pÁ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1)i€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1<pÁ� b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1)i €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1·�Lp  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×LLÅÅ):*:4¯4¯>"?"I•I•ST^…^…h÷h÷risi} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡üüqqææ[ ["Ð"Ð$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I"J"L•L•NOQ{Q{SîSîUaVaXÔXÔZG[G]º]º_-`-b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰̉̋?Œ?޲޲�%‘%“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1õ}!1�õ}!1“¾!1 “¾!1)Å!1)Å!1Áç!1€Áç!1_("1ü_("1û."1û."1“Q"1p“Q"11’"1ì1’"1͘"1ô͘"1«Ø"1`«Ø"1 ý"1Ì ý"10#1ì0#1Ý~#1L Ý~#1·$1¬ ·$1ßð$1p ßð$1½À%1t ½À%1·€&1T ·€&1­P'1˜­P'1Uc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´P b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1•Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´P b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1•Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´P b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1•Õ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1«§  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` × ×NNÍÍ)D*D4»4»>0?0I¥I¥ST^•^•hisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``××NNÅÅ < < ³ ³**¡¡��}}ôôk k"â"â$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n1o1q¢q¢stv„v„xõxõzf{f}×}×H€H‚¹‚¹„*…*‡š‡š‰ Š Œ|Œ|ŽíŽí�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„i!1;}!1;}!1Ù½!1|Ù½!1=Ã!1„=Ã!1sÆ!1ðsÆ!1ýÏ!1|ýÏ!13Ó!1H3Ó!1 ç!1Ô ç!1«'"1P«'"1-"1X-"1E0"1ÄE0"1Ï9"1PÏ9"1="1="1ßP"1¨ßP"1}‘"1$ }‘"1á–"1, á–"1š"1˜ š"1¡£"1$ ¡£"1צ"1ð צ"1Ô"1| Ô"1IÚ"1XIÚ"1©þ"1Ô©þ"1ë+#1ôë+#1:#1Ð:#1Ûn#1\Ûn#1�#1€�#1˜#1¬˜#1Ï$18Ï$1§ò$1ü§ò$1áÞ%1áÞ%1[ì%1à[ì%1)�&1l)�&1ùR'1°ùR'1¡e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¥q b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1­í bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¥q b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1­í  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¥q  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1­í bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‘±­Í2R bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations15uB‚ÄHˆÌ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d5µB„ ŠÆFß_ˆ´4JÊ ŒÎN�TÔ˜Ü\  d ä ( ¨ ì l ° 0 r ² ô 4 x ¸ … Å ü < bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10@@ Ù ÙRRÕÕ)N*N4Ç4Ç>@?@I·I·S,T,^«^«his‘s‘} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð@@¹¹22«« $ $ ½ ½66¯¯((¡¡““  … …"þ"þ$�%�'ú'ú)s*s,ì,ì.e/e1Þ1Þ3W4W6Ð6Ð8I9I;Â;Â=;>;@´@´B-C-E¦E¦GHJ˜J˜LMO†O†QýQýStTtVëVëXbYb[Ù[Ù]P^P`Ç`Çb3?3I§I§ST^›^›hisƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯))££ — —"#%‰%‰'ÿ'ÿ)u*u,ë,ë.a/a1×1×3M4M6Ã6Ã8999;¯;¯=>@{@{BñBñDgEgGÝGÝIsJsLçLçN[O[QÏQÏSCTCV·V·X+Y+[Ÿ[Ÿ]^`‡`‡bûbûdoeogãgãiWjWlËlËnKoKq¿q¿s3t3v§v§xy{�{�}~€w€w‚ë‚ë„_…_‡Ó‡Ó‰GŠGŒ»Œ»Ž/�/‘£‘£“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„i!1;}!1;}!1Ù½!1|Ù½!1=Ã!1„=Ã!1sÆ!1ðsÆ!1ýÏ!1|ýÏ!13Ó!1H3Ó!1ç!1Ôç!1¥'"1P¥'"1-"1X-"1E0"1ÄE0"1Ï9"1PÏ9"1="1="1ÙP"1¨ÙP"1w‘"1$ w‘"1á–"1, á–"1š"1˜ š"1¡£"1$ ¡£"1צ"1ð צ"1Ô"1| Ô"1aÚ"1XaÚ"1Áþ"1ÔÁþ"1,#1ô,#1Ù:#1ÐÙ:#1•o#1\•o#1Å�#1€Å�#1Û#1¬Û#1·$18·$1�ò$1ü�ò$1ÉÞ%1ÉÞ%1Cì%1àCì%1·€&1l·€&1‡R'1°‡R'1/e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¢n b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1­í bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¢n b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1­í  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¢n  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1­í bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ž®­ÍLl bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1s³Ê &f×ˆÈ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dsófæYÙL̉ ?¿^Þ2²3³%¥ˆ˜ ‰úz ë k Ü \ Í M ¾ > ¯ /Z  “ÓF†÷7¨è bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10?? × ×NNÍÍ)D*D4»4»>2?2I§I§ST^—^—hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð??¶¶--¤¤   ³ ³**¡¡��}}ôôk k"â"â$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbêbêd]e]gÐgÐiCjCl¶l¶n3o3q¤q¤stv†v†x÷x÷zh{h}Ù}ÙJ€J‚»‚»„,…,‡ž‡ž‰ŠŒ€Œ€ŽñŽñ�b‘b“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxi!1~!1è~!1¥¾!1d¥¾!1MÅ!1lMÅ!1×Î!1è×Î!1 Ò!1´ Ò!1Óç!1@Óç!1q("1¼q("1/"1Ä/"1©8"1@©8"1ß;"1 ß;"1¥Q"1˜¥Q"1C’"1C’"1ñ˜"1 ñ˜"1{¢"1˜ {¢"1±¥"1d ±¥"1óÒ"1ð óÒ"1UÚ"1Ì UÚ"1µþ"1H µþ"1÷+#1h÷+#1:#1D:#1Ãn#1ÐÃn#1¡�#1ô¡�#1Ÿ˜#1 Ÿ˜#1«$1¬«$1ƒò$1pƒò$1½Þ%1t½Þ%17ì%1T7ì%1�€&1à�€&1QR'1$QR'1ùd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎh4 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 B‚¦æ¿ÿ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎh4 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 B‚¦æ¿ÿ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎh4  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 B‚¦æ¿ÿ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‹«<\1Q bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1=}ºúØ7w´ô  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d=½úz·7tô1±©)în„«+_ßhè%¥:ºâbŸ\Ü ™ Ö V “  P Ð  M Š Ê  G „ Ä ˆÈ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSËË)?*?4’4’>'?'I™I™S T ^‰^‰hûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=Ò=Ò?F@FBºBºD.E.G¢G¢I7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnñnñpcqcsÕsÕuGvGx¹x¹z+{+}�}�€‚�‚�„ó„ó†e‡e‰׉׋IŒI޻޻�-‘-“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1e}!1èe}!1¾!1d¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!17ç!1@7ç!1Õ'"1¼Õ'"11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1 Q"1˜ Q"1§‘"1§‘"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1IÚ"1Ì IÚ"1©þ"1H ©þ"1ë+#1hë+#1 :#1D :#1Én#1ÐÉn#1½Œ#1ô½Œ#1»—#1 »—#1É$1¬É$1¡ò$1p¡ò$1ÛÞ%1tÛÞ%1Uì%1TUì%1/�&1à/�&1ÿR'1$ÿR'1§e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÆ’ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 4t=}È bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÆ’ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 4t=}È bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÆ’  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 4t=}È  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �°­Í0P bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1.n=}ºú7w´ô  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d.®=½úz…·7Ü\tô³31±în«+hè%¥âbŸ\Ü ™ Ö V “  P Ð  M Š Ê  G „ Ä ž Þ  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ø ØPPÈÈ)@*@4¸4¸>?I¦I¦ST^œ^œhisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ØØPPÈÈ @ @ ¸ ¸00¨¨  ˜˜ˆˆ x x"ð"ð$h%h'à'à)X*X,Ð,Ð.H/H1À1À38486°6°8(9(; ; =>@�@�BCE€E€GøGøIPJPLÆLÆN?I’I’ST^z^zhÍhÍr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?K@KB¾B¾D1E1G¤G¤IJLŠLŠNýNýPpQpSãSãUVVVXÉXÉZ<[<]¯]¯_"`"b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢zõzõ|h}hÛÛ�N‚N„Á„Á†T‡T‰ƉƋ8Œ8ŽªŽª�‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1e}!1èe}!1¾!1d¾!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!17ç!1@7ç!1Õ'"1¼Õ'"1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1 Q"1˜ Q"1§‘"1§‘"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"1mÚ"1Ì mÚ"1Íþ"1H Íþ"1,#1h,#1 :#1D :#1Én#1ÐÉn#1S�#1ôS�#1Q˜#1 Q˜#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1·€&1à·€&1‡R'1$‡R'1/e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÄ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 3s=}Ç bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÄ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 3s=}Ç bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1=qÄ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 3s=}Ç  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‘±­ÍRr bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1=}ºúÅ7w´ô  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d=½úz·7tô1±Š înaá«+8¸hè�%¥âbŸ\Ü ™ Ö V “  P Ð  M Š Ê  G „ Ä 5u bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ö ÖLLÂÂ)8*84®4®>?I˜I˜S T ^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††üürrèè^ ^"Ô"Ô$J%J'À'À)6*6,¬,¬."/"1˜1˜346„6„8ú8ú:p;p=æ=æ?\@\BÒBÒDHEHG¾G¾IJL‡L‡NûNûPoQoSãSãUxVxXìXìZ`[`]Ô]Ô_H`Hb¼b¼d0e0g¤g¤ijlŒlŒnoq|q|sðsðudvdxØxØzL{L}À}À4€4‚¨‚¨„…‡�‡�‰ŠŒxŒxŽìŽì�`‘`“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1û}!1èû}!1™¾!1d™¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1Íç!1@Íç!1k("1¼k("11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1ŸQ"1˜ŸQ"1=’"1=’"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1OÚ"1Ì OÚ"1¯þ"1H ¯þ"1ñ+#1hñ+#1‘:#1D‘:#1Mo#1ÐMo#19�#1ô9�#17š#1 7š#1�$1¬�$1eò$1peò$1ŸÞ%1tŸÞ%1ì%1Tì%1€&1à€&1ßQ'1$ßQ'1‡d'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ã]) b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M�“ÓÊ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ã]) b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M�“ÓÊ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ã])  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M�“ÓÊ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ­ÍŒ¬7W bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1_ŸÑþ>›Û8x  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d_ß>¾�ü|œÛ[uõº:PЙ)©xøUÕ›Û2² � ì l É I ¦ & ƒ  ` à ?Þ{»X bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØKK¾¾)1*14¤4¤>?I†I†SûSû]l^lh¼h¼rNsN} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆûûnnááT T"Ç"Ç$:%:'­'­) * ,“,“./1y1y3ì3ì5_6_8Ò8Ò:E;E=¸=¸?)@)BšBšD E G|G|IíIíK^L^NÏNÏP@Q@S±S±U"V"X“X“Z[]u]u_æ_æa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTz¤z¤|}††�÷�÷ƒh„h†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1k}!1Ðk}!1 ¾!1L ¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1=ç!1(=ç!1Û'"1¤Û'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1Q"1€Q"1­‘"1ü­‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#1{2#1,{2#1ÕŒ#1˜ÕŒ#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1í€&1Ìí€&1½R'1½R'1ee'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÇ“ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T=}Ó bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÇ“ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T=}Ó bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÇ“  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T=}Ó  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ­Í�¯3S bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1%e€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1%¥þ~×W°0µ õ  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` × ×NNÅÅ)<*<4³4³>*?*IŸIŸST^‰^‰hþhþrsss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``××NNÅÅ < < ³ ³**¡¡��}}ôôk k"â"â$Y%Y'Ð'Ð)G*G,¾,¾.5/51¬1¬3#4#6š6š89;ˆ;ˆ=ÿ=ÿ?v@vBíBíDdEdGÛGÛIRJRLÇLÇN'?'I›I›ST^ƒ^ƒh÷h÷rmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��‡‡üüq q"æ"æ$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡�‡�‰ŠŒwŒwŽìŽì�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1k}!1Ðk}!1 ¾!1L ¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!17ç!1(7ç!1Õ'"1¤Õ'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1 Q"1€ Q"1§‘"1ü§‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#15#1,5#1q�#1˜q�#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÅ‘ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U=}Ò bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÅ‘ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U=}Ò bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÅ‘  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U=}Ò  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ­Í”´Xx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¾þ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1‰ bâ;»”NŽ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×MMÂÂ)7*74¬4¬>!?!I”I”ST^z^zhíhír`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b£b£deg‰g‰iüiükolonânâpUqUsÈsÈu;v;x®x®z!{!}”}”€‚z‚z„í„í†`‡`‰Ó‰Ó‹FŒF޹޹�,‘,“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1õ}!1Ðõ}!1“¾!1L“¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Çç!1(Çç!1e("1¤e("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#15#1,5#1S�#1˜S�#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1u€&1Ìu€&1ER'1ER'1íd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸R b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 X˜ÚÅ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸R b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 X˜ÚÅ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸R  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 X˜ÚÅ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ­Í�­9Y bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1‡^Þ5µ Œ2r €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84­4­>"?"I•I•ST^y^yhëhër_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?N@NBÃBÃD8E8G­G­I"J"L•L•NOQ{Q{SîSîUaVaXÔXÔZG[G]º]º_-`-bŸbŸdegƒgƒiõiõkglgnÙnÙpKqKs½s½u/v/x¡x¡z{}…}…÷÷�i‚i„Û„Û†O‡O‰‰‹5Œ5ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1q}!1Ðq}!1¾!1L¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Cç!1(Cç!1á'"1¤á'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1Q"1€Q"1³‘"1ü³‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1[Ú"1´ [Ú"1»þ"10 »þ"1ý+#1Pý+#1u2#1,u2#1óŒ#1˜óŒ#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÈ” b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 )i?à  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÈ” b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 )i?à  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÈ”  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 )i?à   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Œ¬2R3S bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1\€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1œ÷wÒR­-Ì  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Þ ÞTTÊÊ)@*@4¶4¶>*?*IžIžST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÞÞXXÒÒ L L Æ Æ@@ºº44®®$$šš † †"ü"ü$r%r'è'è)^*^,Ô,Ô.J/J1À1À36466¬6¬8"9";˜;˜=>@‚@‚BöBöDjEjGÞGÞIRJRLÆLÆN:O:Q®Q®S"T"V–V–X Y [~[~]ò]ò_f`fbÚbÚdNeNgÂgÂi6j6lªlªnoq’q’stvzvzxîxîzb{b}Ö}ÖJ€J‚¾‚¾„2…2‡¦‡¦‰ŠŒŒŒŒŽÿŽÿ�r‘r“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1ï}!1Ðï}!1�¾!1L�¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1»ç!1(»ç!1Y("1¤Y("1[/"1¬[/"1å8"1(å8"1<"1ô<"1�Q"1€�Q"1+’"1ü+’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1u2#1,u2#1“Œ#1˜“Œ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1ç€&1Ìç€&1·R'1·R'1_e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬E b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 b¢“Óî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬E b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 b¢“Óî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬E  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 b¢“Óî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ­Í¦Æ\| bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1^€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1žósÈH�n ®  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>1?1I¨I¨ST^‘^‘hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIYJYLÐLÐNGOGQ¾Q¾S5T5V¬V¬X#Y#[š[š]^`ˆ`ˆbÿbÿdvevgígíidjdlÛlÛnMoMq¿q¿s1t1v£v£xy{‡{‡}ù}ùk€k‚݂݄O…O‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1q}!1Ðq}!1¾!1L¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1=ç!1(=ç!1Û'"1¤Û'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1Q"1€Q"1­‘"1ü­‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#15#1,5#1��#1˜��#1Ÿ$1$Ÿ$1wò$1èwò$1±Þ%1ì±Þ%1]€&1Ì]€&1-R'1-R'1Õd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÆ’ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 (h>~Þ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÆ’ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 (h>~Þ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÆ’  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 (h>~Þ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 —·/^~ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1·÷€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ˆcã>¾™g§ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜPPÄÄ)8*84¬4¬>?I�I�ST^t^thæhærXsX} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœ„„øøl l"à"à$T%T'È'È)<*<,°,°.$/$1˜1˜3 4 6€6€8ô8ô:h;h=Ü=Ü?N@NBÀBÀD2E2G¤G¤IJLˆLˆNúNúPlQlSÞSÞUPVPXÂXÂZ4[4]¦]¦_`bŠbŠdüdüfngniàiàkRlRnÄnÄp6q6s¨s¨uvxŒxŒzþzþ|p}pââ�T‚T„ƄƆ8‡8‰©‰©‹ŒŽ‹Ž‹�ü�ü’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1ï}!1Ðï}!1�¾!1L�¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Áç!1(Áç!1_("1¤_("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1“Q"1€“Q"11’"1ü1’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#15#1,5#1)�#1˜)�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­G b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 c£“Óð0 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­G b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 c£“Óð0 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­G  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 c£“Óð0  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ­Í†¦+K bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1³ó€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ròGÇœñqC €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ø ØMMÂÂ)7*74¬4¬>#?#I˜I˜S T ^^hñhñrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccØØMM 7 7 ¬ ¬!!––  €€õõjjßßT T"É"É$>%>'³'³)(*(,�,�./1‡1‡3ü3ü5q6q8æ8æ:[;[=Ð=Ð?E@EBºBºD/E/G¤G¤IJL�L�NOQzQzSïSïUdVdXÙXÙZN[N]Ã]Ã_8`8b­b­d"e"g—g—i j l�l�nónópeqes×s×uIvIx»x»z-{-}Ÿ}Ÿ€‚ƒ‚ƒ„õ„õ†g‡g‰ىًKŒK޽޽�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1w}!1Ðw}!1¾!1L¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1Cç!1(Cç!1á'"1¤á'"1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1Q"1€Q"1³‘"1ü³‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1o2#1,o2#1 �#1˜ �#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1#�&1Ì#�&1óR'1óR'1›e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÉ• b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IU•@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÉ• b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IU•@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÉ•  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IU•@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ÿ?Qq bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1S€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1“ðpÍMª*ã # €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ý ÝVVÏÏ)H*H4Á4Á>:?:I±I±S&T&^™^™h i ss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÝÝVVÏÏ H H Á Á::³³,,¥¥—— ‰ ‰"#%{%{'ô'ô)m*m,æ,æ._/_1Ø1Ø3Q4Q6Ê6Ê8C9C;¼;¼=5>5@®@®B'C'E E GHJ’J’L M O€O€Q÷Q÷SnTnVåVåX\Y\[Ó[Ó]J^J`Á`Áb6c6e«e«g h j•j•l m o}o}qðqðsctcvÖvÖxIyI{¼{¼}/~/€¢€¢‚ƒ…ˆ…ˆ‡û‡û‰mŠmŒߌߎQ�Q‘ÑÓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1é}!1Ðé}!1‡¾!1L‡¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1µç!1(µç!1S("1¤S("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1‡Q"1€‡Q"1%’"1ü%’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1mÚ"1´ mÚ"1Íþ"10 Íþ"1,#1P,#1o2#1,o2#1iŒ#1˜iŒ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1ù€&1Ìù€&1ÉR'1ÉR'1qe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡: b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 b¢¸øâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡: b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 b¢¸øâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡:  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 b¢¸øâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ­Í›»Ff bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Q€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1‘äd·7Š A �  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii ã ã]]ÓÓ)I*I4¿4¿>5?5I©I©ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiãã]]×× Q Q Ë ËEE¿¿99³³--§§!! › ›"#%‹%‹'(*w*w,í,í.c/c1Ù1Ù3O4O6Å6Å8;9;;±;±='>'@�@�BCE‰E‰GÿGÿIuJuLéLéN]O]QÑQÑSETEV¹V¹X-Y-[¡[¡]^`‰`‰býbýdqeqgågåiYjYlÍlÍnAoAqµqµs)t)v�v�xy{…{…}ù}ùm€m‚á‚á„U…U‡ɇɉ=Š=Œ±Œ±Ž%�%‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1q}!1Ðq}!1¾!1L¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1Cç!1(Cç!1á'"1¤á'"1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1Q"1€Q"1³‘"1ü³‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#15#1,5#1§�#1˜§�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1·€&1Ì·€&1‡R'1‡R'1/e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÇ“ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JT”þ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÇ“ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JT”þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÇ“  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JT”þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �°2RPp bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1°ð€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1‡däAÁž€À €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I£I£ST^‡^‡høhørjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbæbædYeYgÌgÌi?j?l²l²n#o#q”q”stvvvvxçxçzX{X}É}É:€:‚«‚«„…‡އމÿ‰ÿ‹pŒpŽáŽá�R‘R“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1é}!1Ðé}!1‡¾!1L‡¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1»ç!1(»ç!1Y("1¤Y("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1�Q"1€�Q"1+’"1ü+’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1=Ú"1´ =Ú"1�þ"10 �þ"1ß+#1Pß+#15#1,5#1ÿŒ#1˜ÿŒ#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢< b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 c£¹ùä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢< b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 c£¹ùä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢<  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 c£¹ùä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ­Í{›9 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¤ä€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1]Ý0°ƒÖVÔ  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßWWËË)?*?4³4³>'?'I™I™S T ^}^}hïhïrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ��  "÷"÷$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1}}!1Ð}}!1¾!1L¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1Iç!1(Iç!1ç'"1¤ç'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1Q"1€Q"1¹‘"1ü¹‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1i2#1,i2#1)�#1˜)�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1;�&1Ì;�&1 S'1 S'1³e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÊ– b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ^=}É  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÊ– b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ^=}É  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÊ–  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ^=}É   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 r’ )Ll bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 J€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ŠéiÈH§'ú : €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜTTÌÌ)D*D4¼4¼>4?4IªIªS T ^–^–h i s‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœŒŒ | |"ô"ô$l%l'ä'ä)\*\,Ô,Ô.L/L1Ä1Ä3<4<6´6´8,9,;¤;¤=>@”@”B C E„E„GüGüItJtLêLêN`O`QÖQÖSLTLVÂVÂX8Y8[®[®]$^$`š`šbce†e†gügüirjrlèlèn^o^qÔqÔsJtJvÀvÀx6y6{¬{¬}"~"€˜€˜‚ƒ…„…„‡ú‡ú‰pŠpŒæŒæŽ\�\‘Ò‘Ò“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1Ý}!1ÐÝ}!1{¾!1L{¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1¯ç!1(¯ç!1M("1¤M("1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1�Q"1€�Q"1’"1ü’"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1c2#1,c2#1?Œ#1˜?Œ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–/û b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹È—× bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–/û b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹È—× bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–/û  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹È—×  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �°Rr0P bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1„ÕU¦&w÷ T  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜQQÆÆ);*;4°4°>#?#I–I–S T ^|^|hïhïrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒvvëë` `"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?O@OBÂBÂD5E5G¨G¨IJLŽLŽNOQtQtSçSçUZVZXÍXÍZ@[@]³]³_&`&b™b™d e ggiòiòkelenØnØpKqKs¾s¾u1v1x¤x¤z{}Š}Šýý�p‚p„ã„ã†V‡V‰ȉȋ:Œ:ެެ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1w}!1Ðw}!1¾!1L¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Iç!1(Iç!1ç'"1¤ç'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1Q"1€Q"1¹‘"1ü¹‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1 5#1, 5#1Å�#1˜Å�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÈ” b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]=}Ê  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÈ” b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]=}Ê  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÈ”  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]=}Ê   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‡§Cc>^ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1©é€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1†eåDÄ#£™Ù €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÆÆ)<*<4²4²>(?(IœIœST^„^„høhørlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$N%N'Ä'Ä):*:,°,°.&/&1œ1œ346ˆ6ˆ8þ8þ:t;t=ê=ê?`@`BÖBÖDLELGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡Œ‡Œ‰ŠŒtŒtŽèŽè�\‘\“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1ã}!1Ðã}!1�¾!1L�¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1¯ç!1(¯ç!1M("1¤M("1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1�Q"1€�Q"1’"1ü’"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1mÚ"1´ mÚ"1Íþ"10 Íþ"1,#1P,#1÷4#1,÷4#1ÕŒ#1˜ÕŒ#1Ÿ$1$Ÿ$1wò$1èwò$1±Þ%1ì±Þ%1!€&1Ì!€&1ñQ'1ñQ'1™d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 JŠÇ˜Ø bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 JŠÇ˜Ø bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 JŠÇ˜Ø  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p�'?_ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1•Õ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1HÈ™êj»;¥ å  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜOOÂÂ)5*54¨4¨>?IŠIŠSûSû]l^lhÝhÝrNsN} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒÿÿrrååX X"Ë"Ë$>%>'±'±)$*$,—,—. / 1}1}3ð3ð5c6c8Ö8Ö:I;I=¼=¼?-@-BžBžDEG€G€IñIñKbLbNÓNÓPDQDSµSµU&V&X—X—Z[]y]y_ê_êa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1}}!1Ð}}!1¾!1L¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Oç!1(Oç!1í'"1¤í'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1!Q"1€!Q"1¿‘"1ü¿‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1i2#1,i2#1M�#1˜M�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ÿ€&1Ìÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvË— b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @>~Í  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvË— b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @>~Í  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvË—  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @>~Í   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 e…óGg bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1B€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1‚ãcÄD¥%Z €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I£I£ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbèbèd]e]gÒgÒiGjGl¼l¼n1o1q¦q¦stv�v�xy{z{z}ï}ïd€d‚قلN…N‡ÇÉ8Š8Œ­Œ­Ž"�"‘—‘—“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1×}!1Ð×}!1u¾!1Lu¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1©ç!1(©ç!1G("1¤G("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1{Q"1€{Q"1’"1ü’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1c2#1,c2#1ŸŒ#1˜ŸŒ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹$ð b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 NŽË ¯ï bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹$ð b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 NŽË ¯ï bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹$ð  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 NŽË ¯ï  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 …¥:[{ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ØX°0ˆp °  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áVVËË)@*@4µ4µ>+?+IŸIŸST^‡^‡hûhûrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‹‹ u u"ê"ê$_%_'Ô'Ô)I*I,¾,¾.3/31¨1¨346’6’89;|;|=ñ=ñ?f@fBÛBÛDPEPGÅGÅI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÃbÃd7e7g«g«ijl“l“noq{q{sïsïucvcx×x×zK{K}¿}¿3€3‚§‚§„…‡‘‡‘‰ŠŒ{Œ{ŽðŽð�e‘e“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1}}!1Ð}}!1¾!1L¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Oç!1(Oç!1í'"1¤í'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1!Q"1€!Q"1¿‘"1ü¿‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1 5#1, 5#1ï�#1˜ï�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÉ• b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @>~Î bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÉ• b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @>~Î bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÉ•  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @>~Î  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~žBb,L bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1£ã€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1†gçHÈ)©»û €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÆÆ);*;4°4°>%?%I˜I˜S T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnænæpYqYsÌsÌu?v?x²x²z%{%}˜}˜ € ‚~‚~„ñ„ñ†d‡d‰׉׋JŒJ޽޽�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1Ý}!1ÐÝ}!1{¾!1L{¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1©ç!1(©ç!1G("1¤G("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1{Q"1€{Q"1’"1ü’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1ý4#1,ý4#1;�#1˜;�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ&ò b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M�Ê ¯ï bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ&ò b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M�Ê ¯ï bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ&ò  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M�Ê ¯ï  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 `€ðZz bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1˜Ø€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`à8¸�èhH €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝRRÇÇ)<*<4±4±>&?&I™I™S T ^}^}hïhïrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��wwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD.?.I¢I¢ST^Š^Šhþhþrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²((žž Š Š"#%v%v'ì'ì)b*b,Ø,Ø.N/N1Ä1Ä3:4:6°6°8&9&;œ;œ=>@†@†BúBúDnEnGâGâIVJVLÊLÊN>O>Q²Q²S&T&VšVšXY[‚[‚]ö]ö_j`jbÞbÞdReRgÆgÆi:j:l®l®n"o"q–q–s t v~v~xòxòzf{f}Ú}ÚN€N‚‚„6…6‡ª‡ª‰ŠŒ�Œ�Ž�‘v‘v“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1Ñ}!1ÐÑ}!1o¾!1Lo¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1£ç!1(£ç!1A("1¤A("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1uQ"1€uQ"1’"1ü’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1i2#1,i2#1S�#1˜S�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1ù€&1Ìù€&1ÉR'1ÉR'1qe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€å b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘΢â bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€å b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘΢â bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€å  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘뢉  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 zš$Tt bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ãcÆF©) ` €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii à àWWÎÎ)E*E4¼4¼>5?5I¬I¬S#T#^•^•hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiààWWÎÎ E E ¼ ¼33ªª!!˜˜††ýýt t"ë"ë$b%b'Ù'Ù)P*P,Ç,Ç.>/>1µ1µ3,4,6£6£89;‘;‘=>@@BöBöDmEmGäGäI]J]LÔLÔNKOKQÂQÂS9T9V°V°X'Y'[ž[ž]^`Œ`ŒbcezezgñgñihjhlßlßnQoQqÃqÃs5t5v§v§xy{‹{‹}ý}ýo€o‚á‚á„S…S‡Ňʼn7Š7Œ©Œ©Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1ƒ}!1Ѓ}!1!¾!1L!¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Uç!1(Uç!1ó'"1¤ó'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1'Q"1€'Q"1Å‘"1üÅ‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#15#1,5#1[Ž#1˜[Ž#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1o€&1Ìo€&1?R'1?R'1çd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwÊ– b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R=}Ð bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwÊ– b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R=}Ð bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwÊ–  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R=}Ð  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ­Íu•: bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1§ç€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1œƒjêQÑ'g €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àTTÈÈ)<*<4°4°>"?"I”I”ST^x^xhêhêr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  ˆˆüüp p"ä"ä$X%X'Ì'Ì)@*@,´,´.(/(1œ1œ346„6„8ø8ø:l;l=à=à?R@RBÄBÄD6E6G¨G¨IJLŒLŒNþNþPpQpSâSâUTVTXÆXÆZ8[8]ª]ª_`bŽbŽdegrgriäiäkVlVnÈnÈp:q:s¬s¬uvx�x�z{}t}tææ�X‚X„ʄʆ<‡<‰­‰­‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1Ñ}!1ÐÑ}!1o¾!1Lo¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1£ç!1(£ç!1A("1¤A("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1uQ"1€uQ"1’"1ü’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#15#1,5#1û�#1˜û�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Ÿ€&1ÌŸ€&1oR'1oR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ç b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 P�Í £ã bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ç b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 P�Í £ã bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ç  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 P�Í £ã  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Yyâ[{ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1£ã€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Œ oïRÒ5µÃ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ü ÜQQÆÆ);*;4°4°>'?'IœIœST^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÜÜQQÆÆ ; ; ° °%%šš„„ùùnnããX X"Í"Í$B%B'·'·),*,,¡,¡./1‹1‹346u6u8ê8ê:_;_=Ô=Ô?I@IB¾B¾D3E3G¨G¨IJL”L”N O Q~Q~SóSóUhVhXÝXÝZR[R]Ç]Ç_<`M?MIÄIÄS9T9^¬^¬his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðyyòòkkää ] ] Ö ÖOOÈÈAAºº33¬¬%% ž ž"#%�%�' ( *‚*‚,û,û.t/t1í1í3f4f6ß6ß8X9X;Ñ;Ñ=J>J@Ã@ÃB@?@I´I´S(T(^œ^œhis„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvððjjää ^ ^ Ø ØRRÌÌFFÀÀ::´´.. ¨ ¨""#"%˜%˜'(*„*„,ú,ú.p/p1æ1æ3\4\6Ò6Ò8H9H;¾;¾=4>4@ª@ªB C E–E–G H J€J€LôLôNhOhQÜQÜSPTPVÄVÄX8Y8[¬[¬] ^ `”`”bce|e|gðgðidjdlØlØnLoLqÀqÀs4t4v¨v¨xy{�{�}~€x€x‚ì‚ì„`…`‡Ô‡Ô‰HŠHŒ¼Œ¼Ž0�0‘¤‘¤“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1×}!1�×}!1u¾!1u¾!1­Â!1$­Â!1©ç!1�©ç!1G("1G("1,"1$,"1{Q"1�{Q"1’"1’"1Q–"1$Q–"1íØ"1�íØ"1Mý"1üMý"1)0#1 )0#1õ~#1| õ~#1Û$1Ü Û$1³ò$1  ³ò$1íÞ%1¤ íÞ%1Õ€&1„ Õ€&1¥R'1È¥R'1Me'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations16jS‡Ú¦ê b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations16jS‡Ú¦ê b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations16jS‡Ú¦ê b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1­Í  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww î îeeÜÜ)S*S4Ê4Ê>???I´I´S'T'^˜^˜h i sysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwîîeeÜÜ S S Ê ÊAA¸¸//¦¦””  ‚ ‚"ù"ù$p%p'ç'ç)^*^,Õ,Õ.L/L1Ã1Ã3:4:6±6±8(9(;Ÿ;Ÿ=>@�@�BCE{E{GòGòIgJgLÜLÜNQOQQÆQÆS;T;V°V°X%Y%[š[š]^`„`„b÷b÷djejgÝgÝiPjPlÃlÃn4o4q¥q¥stv‡v‡xøxøzi{i}Ú}ÚK€K‚¼‚¼„-…-‡�‡�‰ŠŒŒŽðŽð�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ë}!1�Ë}!1i¾!1 i¾!1kÅ!1kÅ!1�ç!1€�ç!1;("1ü;("1=/"1=/"1oQ"1poQ"1 ’"1ì ’"1™"1ô™"1‘Ú"1`‘Ú"1ñþ"1Ìñþ"1)0#1ì)0#1õ~#1L õ~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s Ù b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1»û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s Ù b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1»û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s Ù b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1»û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Qq  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv î îffÚÚ)N*N4Â4Â>4?4I¦I¦ST^Š^Šhühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvîîffÞÞ V V Î ÎFF¾¾66®®&&žž Ž Ž"#%z%z'î'î)b*b,Ö,Ö.J/J1¾1¾32426¦6¦89;Ž;Ž=>@v@vBêBêD^E^GÒGÒIDJDL¶L¶N(O(QšQšS T V~V~XðXðZb[b]Ô]Ô_F`Fb¸b¸d*e*gœgœijl€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1•}!1�•}!13¾!1 3¾!1¿Â!1¿Â!1gç!1€gç!1("1ü("1‘,"1‘,"19Q"1p9Q"1ב"1ìב"1c–"1ôc–"1çØ"1`çØ"1Gý"1ÌGý"1)0#1ì)0#1õ~#1L õ~#1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1S�&1T S�&1#S'1˜#S'1Ëe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†Û§ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1S€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†Û§ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1S€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†Û§ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1S €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1­Í  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu í íeeÝÝ)U*U4Í4Í>C?CI¹I¹S/T/^¥^¥his‘s‘} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuííeeÝÝ U U Í ÍEE½½55­­%%�� � �"#%}%}'õ'õ)m*m,å,å.]/]1Õ1Õ3M4M6Å6Å8=9=;µ;µ=->-@¥@¥BCE•E•G H JƒJƒLùLùNoOoQåQåS[T[VÑVÑXGYG[½[½]3^3`©`©bce•e•g h j�j�l÷l÷nmomqãqãsYtYvÏvÏxEyE{»{»}1~1€§€§‚ƒ…“…“‡ ˆ ŠŠŒõŒõŽk�k‘á‘á“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ñ}!1�Ñ}!1o¾!1 o¾!1MÅ!1MÅ!1£ç!1€£ç!1A("1üA("1/"1/"1uQ"1puQ"1’"1ì’"1ñ˜"1ôñ˜"1çØ"1`çØ"1Gý"1ÌGý"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}â b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¢â€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}â b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¢â€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}â b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¢â €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1­Í  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo å åZZÏÏ)D*D4·4·>*?*I�I�ST^ƒ^ƒhöhörgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooåå[[ÑÑ G G ½ ½33©©••  ôôi i"Þ"Þ$S%S'È'È)=*=,²,².'/'1œ1œ346†6†8û8û:p;p=ã=ã?V@VBÉBÉD5?5I©I©ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðsséé__ÕÕ K K Á Á77­­##™™……ûûq q"ç"ç$]%]'Ó'Ó)I*I,¿,¿.5/51«1«3!4!6—6—8 9 ;ƒ;ƒ=ù=ù?o@oBåBåD[E[GÑGÑIEJEL¹L¹N-O-Q¡Q¡STV‰V‰XýXýZq[q]å]å_Y`YbÍbÍdAeAgµgµi)j)l�l�noq…q…sùsùumvmxáxázU{U}É}É=€=‚±‚±„%…%‡™‡™‰ Š Œ�Œ�ŽõŽõ�i‘i“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ñ}!1�Ñ}!1o¾!1 o¾!1MÅ!1MÅ!1£ç!1€£ç!1A("1üA("1/"1/"1uQ"1puQ"1’"1ì’"1ñ˜"1ôñ˜"1çØ"1`çØ"1Gý"1ÌGý"1)0#1ì)0#1õ~#1L õ~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%13€&1T 3€&1R'1˜R'1«d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|â b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1£ã€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|â b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1£ã€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|â b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1£ã €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1­Í  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ç çZZÍÍ)@*@4±4±>"?"I“I“ST^u^uhæhærUsU} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqçç]]ÓÓ I I ¿ ¿55««!!——  }}ððc c"Ö"Ö$I%I'¼'¼)/*/,¢,¢./1ˆ1ˆ3û3û5n6n8á8á:T;T=Å=Å?6@6B§B§DEG‰G‰IúIúKkLkNÜNÜPMQMS¾S¾U/V/X X Z[]‚]‚_ó_óadbddÕdÕfFgFi·i·k(l(n™n™p q s{s{uìuìw]x]zÎzÎ|?}?°°�!‚!„’„’†‡‰q‰q‹á‹á�QŽQ�Á�Á’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1�}!1��}!1-¾!1 -¾!1³Â!1³Â!1aç!1€aç!1ÿ'"1üÿ'"1…,"1…,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1W–"1ôW–"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OƒØ¤ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OƒØ¤ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OƒØ¤ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1­Í  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq è è__ÖÖ)M*M4Ä4Ä>9?9I®I®S#T#^˜^˜h i s‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqèè__ÖÖ M M Ä Ä;;²²))  ŽŽ | |"ó"ó$j%j'á'á)X*X,Ï,Ï.F/F1½1½34446«6«8"9";™;™=>@‡@‡BþBþDuEuGìGìIaJaLÖLÖNKOKQÀQÀS5T5VªVªXY[”[”] ^ `~`~bóbódhehgÝgÝiRjRlÇlÇn.?.I¢I¢ST^Š^Šhþhþrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnææ^^ÖÖ N N Æ Æ>>¶¶..¦¦�� z z"ï"ï$d%d'Ù'Ù)N*N,Ã,Ã.8/81­1­3"4"6—6—8 9 ;�;�=ö=ö?k@kBàBàDUEUGÊGÊI>J>L²L²N&O&QšQšSTV‚V‚XöXöZj[j]Þ]Þ_R`RbÆbÆd:e:g®g®i"j"l–l–n o q~q~sòsòufvfxÚxÚzN{N}Â}Â6€6‚ª‚ª„…‡’‡’‰ŠŒ|Œ|ŽñŽñ�f‘f“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1�}!1��}!1-¾!1 -¾!1³Â!1³Â!1[ç!1€[ç!1ù'"1üù'"1…,"1…,"1-Q"1p-Q"1Ë‘"1ìË‘"1W–"1ôW–"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Ô  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Ô  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Ô  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1­Í  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æ\\ÑÑ)F*F4»4»>.?.I¡I¡ST^‡^‡húhúrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  ‚‚øøn n"ä"ä$Y%Y'Î'Î)C*C,¸,¸.-/-1¢1¢346Œ6Œ89;v;v=ë=ë?`@`BÕBÕDJEJG¿G¿I2J2L¥L¥NOQ‹Q‹SþSþUqVqXäXäZW[W]Ê]Ê_=`=b°b°d#e#g–g–i j l|l|nïnïpbqbsÕsÕuHvHx»x»z.{.}¡}¡€‚‡‚‡„ú„ú†m‡m‰à‰à‹SŒSŽÆŽÆ�9‘9“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1×}!1�×}!1u¾!1 u¾!1kÅ!1kÅ!1©ç!1€©ç!1G("1üG("1=/"1=/"1{Q"1p{Q"1’"1ì’"1™"1ô™"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ë b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ë b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ë b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1­Í  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn ä äYYÎÎ)C*C4¸4¸>+?+IžIžST^‚^‚hôhôrfsf} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnääZZÐÐ F F ¼ ¼22¨¨””  ~~óóh h"Ý"Ý$R%R'Ç'Ç)<*<,±,±.&/&1›1›346…6…8ú8ú:o;o=ä=ä?Y@YBÎBÎDCECG¸G¸I+J+LžLžNOQ„Q„S÷S÷UjVjXÝXÝZP[P]Ã]Ã_6`6b¨b¨degŒgŒiþiþkplpnânâpTqTsÆsÆu8v8xªxªz{}Ž}Ž€‚r‚r„ä„ä†V‡V‰ɉɋ<Œ<ޝޝ�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1‰}!1�‰}!1'¾!1 '¾!1ËÂ!1ËÂ!1[ç!1€[ç!1ù'"1üù'"1�,"1�,"1-Q"1p-Q"1Ë‘"1ìË‘"1o–"1ôo–"1mÚ"1`mÚ"1Íþ"1ÌÍþ"10#1ì0#1é~#1L é~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ç€&1T ç€&1·R'1˜·R'1_e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Õ¡ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1*j€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Õ¡ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1*j€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Õ¡ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1*j €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1;[  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ë ëaa××)M*M4Á4Á>5?5I©I©ST^‘^‘hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqëëeeßß Y Y Ó ÓMMÇÇAA»»11§§ “ “" # %%'õ'õ)k*k,á,á.W/W1Í1Í3C4C6¹6¹8/9/;¥;¥=>@�@�BCEuEuGéGéI]J]LÑLÑNEOEQ¹Q¹S-T-V¡V¡XY[‰[‰]ý]ý_q`qbåbådYeYgÍgÍiAjAlµlµn)o)q�q�stv…v…xùxùzm{m}á}áU€U‚ɂɄ=…=‡¯‡¯‰"Š"Œ•Œ•Ž�‘{‘{“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1Ý}!1�Ý}!1{¾!1 {¾!15Å!15Å!1¯ç!1€¯ç!1M("1üM("1/"1/"1�Q"1p�Q"1’"1ì’"1Ù˜"1ôÙ˜"1áØ"1`áØ"1Aý"1ÌAý"10#1ì0#1é~#1L é~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ù€&1T ù€&1ÉR'1˜ÉR'1qe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�(ô b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1“Ó€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�(ô b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1“Ó€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�(ô b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1“Ó €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1­Í  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj á áXXÏÏ)F*F4½4½>4?4I«I«S"T"^”^”hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjááXXÏÏ F F ½ ½44««""™™‡‡þþu u"ì"ì$c%c'Ú'Ú)Q*Q,È,È.?/?1¶1¶3-4-6¤6¤89;’;’= > @€@€B÷B÷DnEnGåGåI\J\LÓLÓNJOJQÁQÁS8T8V¯V¯X&Y&[�[�]^`‹`‹bceyeygðgðigjglÞlÞnPoPqÂqÂs4t4v¦v¦xy{Š{Š}ü}ün€n‚à‚à„R…R‡ćĉ6Š6Œ¨Œ¨Ž�‘Œ‘Œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ƒ}!1�ƒ}!1!¾!1 !¾!1ÑÂ!1ÑÂ!1Uç!1€Uç!1ó'"1üó'"1£,"1£,"1'Q"1p'Q"1Å‘"1ìÅ‘"1u–"1ôu–"1yÚ"1`yÚ"1Ùþ"1ÌÙþ"10#1ì0#1é~#1L é~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1o€&1T o€&1?R'1˜?R'1çd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ñ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1+k€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ñ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1+k€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ñ� b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1+k €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ff  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo ç ç[[ÏÏ)C*C4µ4µ>'?'I™I™S T ^}^}hïhïr_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooçç__×× O O Ç Ç??··//§§�� w w"ë"ë$_%_'Ó'Ó)G*G,»,».///1£1£346‹6‹8ÿ8ÿ:s;s=å=å?W@WBÉBÉD;E;G­G­IJL‘L‘NOQuQuSçSçUYVYXËXËZ=[=]¯]¯_!`!b“b“degwgwiéiék[l[nÍnÍp?q?s±s±u#v#x•x•z{}y}yëë�]‚]„τφ?‡?‰°‰°‹!Œ!Ž’Ž’�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1Ý}!1�Ý}!1{¾!1 {¾!15Å!15Å!1¯ç!1€¯ç!1M("1üM("1/"1/"1�Q"1p�Q"1’"1ì’"1Ù˜"1ôÙ˜"1áØ"1`áØ"1Aý"1ÌAý"10#1ì0#1é~#1L é~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ(ô b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1“Ó€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ(ô b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1“Ó€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ(ô b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1“Ó €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1­Í  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ß ßTTÉÉ)>*>4³4³>(?(I�I�ST^„^„höhörhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjßßTTÉÉ > > ³ ³((��‡‡üüqqææ[ ["Ð"Ð$E%E'º'º)/*/,¤,¤./1Ž1Ž346x6x8í8í:b;b=×=×?L@LBÁBÁD6E6G«G«I J L•L•N O QQSôSôUiViXÞXÞZS[S]È]È_=`=b²b²d'e'gœgœijl†l†nønøpjqjsÜsÜuNvNxÀxÀz2{2}¤}¤€‚ˆ‚ˆ„ú„ú†l‡l‰Þ‰Þ‹PŒPŽÂŽÂ�4‘4“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ƒ}!1�ƒ}!1!¾!1 !¾!1³Â!1³Â!1Uç!1€Uç!1ó'"1üó'"1…,"1…,"1'Q"1p'Q"1Å‘"1ìÅ‘"1W–"1ôW–"1mÚ"1`mÚ"1Íþ"1ÌÍþ"10#1ì0#1é~#1L é~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1/�&1T /�&1ÿR'1˜ÿR'1§e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Òž b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Òž b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Òž b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1>^  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm æ æ__ØØ)Q*Q4Ê4Ê>A?AI¸I¸S-T-^ ^ his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmææ__ØØ Q Q Ê ÊCC¼¼55®®''   ’ ’" # %„%„'ý'ý)v*v,ï,ï.h/h1á1á3Z4Z6Ó6Ó8L9L;Å;Å=>>>@·@·B0C0E©E©G"H"J™J™LMO‡O‡QþQþSuTuVìVìXcYc[Ú[Ú]Q^Q`È`Èb=c=e²e²g'h'jœjœlmo„o„q÷q÷sjtjvÝvÝxPyP{Ã{Ã}6~6€©€©‚ƒ…�…�‡ˆŠrŠrŒäŒäŽV�V‘ȑȓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ã}!1�ã}!1�¾!1 �¾!1_Å!1_Å!1¯ç!1€¯ç!1M("1üM("11/"11/"1�Q"1p�Q"1’"1ì’"1™"1ô™"1ÛØ"1`ÛØ"1;ý"1Ì;ý"10#1ì0#1é~#1L é~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜1ý b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1À€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜1ý b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1À€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜1ý b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1À €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1­Í  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ä ä^^ÔÔ)J*J4À4À>4?4I¨I¨ST^�^�hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjää^^ØØ R R Ì ÌFFÀÀ::´´..¨¨"" œ œ"#%Œ%Œ'(*x*x,î,î.d/d1Ú1Ú3P4P6Æ6Æ8<9<;²;²=(>(@ž@žBCEŠEŠGHJtJtLèLèN\O\QÐQÐSDTDV¸V¸X,Y,[ [ ]^`ˆ`ˆbübüdpepgägäiXjXlÌlÌn@o@q´q´s(t(vœvœxy{„{„}ø}øl€l‚à‚à„T…T‡ȇȉ<Š<Œ°Œ°Ž$�$‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1}}!1�}}!1¾!1 ¾!1­Â!1­Â!1Oç!1€Oç!1í'"1üí'"1,"1,"1!Q"1p!Q"1¿‘"1ì¿‘"1Q–"1ôQ–"1Ú"1`Ú"1ßþ"1Ìßþ"10#1ì0#1é~#1L é~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1½€&1T ½€&1�R'1˜�R'15e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Κ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Κ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Κ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ss  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk â âYYÐÐ)G*G4¾4¾>3?3I¨I¨ST^Œ^Œhýhýrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkââYYÐÐ G G ¾ ¾55¬¬##ššˆˆÿÿv v"í"í$d%d'Û'Û)R*R,É,É.@/@1·1·3.4.6¥6¥89;“;“= > @�@�BøBøDoEoGæGæI[J[LÐLÐNEOEQºQºS/T/V¤V¤XY[Ž[Ž]^`x`xbëbëd^e^gÑgÑiDjDl·l·n(o(q™q™s t v{v{xìxìz]{]}Î}Î?€?‚°‚°„!…!‡‘‡‘‰ŠŒsŒsŽäŽä�U‘U“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ã}!1�ã}!1�¾!1 �¾!1_Å!1_Å!1¯ç!1€¯ç!1M("1üM("11/"11/"1�Q"1p�Q"1’"1ì’"1™"1ô™"1ÛØ"1`ÛØ"1;ý"1Ì;ý"10#1ì0#1é~#1L é~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¿ÿ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¿ÿ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¿ÿ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1­Í  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âZZÎÎ)B*B4¶4¶>(?(IšIšS T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢’’  ‚ ‚"ú"ú$n%n'â'â)V*V,Ê,Ê.>/>1²1²3&4&6š6š89;‚;‚=ö=ö?j@jBÞBÞDRERGÆGÆI8J8LªLªNOQŽQŽSTVrVrXäXäZV[V]È]È_:`:b¬b¬deg�g�ijltltnænæpXqXsÊsÊuŒ>ްް�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1}}!1�}}!1¾!1 ¾!1ÅÂ!1ÅÂ!1Iç!1€Iç!1ç'"1üç'"1—,"1—,"1Q"1pQ"1¹‘"1칑"1i–"1ôi–"1aÚ"1`aÚ"1Áþ"1ÌÁþ"10#1ì0#1ã~#1L ã~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%15�&1T 5�&1S'1˜S'1­e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÏ› b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1'g€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÏ› b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1'g€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÏ› b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1'g €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations19Y  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áYYÑÑ)I*I4Á4Á>7?7I­I­S#T#^™^™his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‘‘  � �"ù"ù$q%q'é'é)a*a,Ù,Ù.Q/Q1É1É3A4A6¹6¹8191;©;©=!>!@™@™BCE‰E‰GHJwJwLíLíNcOcQÙQÙSOTOVÅVÅX;Y;[±[±]'^'`�`�bce‰e‰gÿgÿiujulëlënaoaq×q×sMtMvÃvÃx9y9{¯{¯}%~%€›€›‚ƒ…‡…‡‡ý‡ý‰sŠsŒéŒéŽ_�_‘Õ‘Õ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1é}!1�é}!1‡¾!1 ‡¾!1/Å!1/Å!1µç!1€µç!1S("1üS("1/"1/"1‡Q"1p‡Q"1%’"1ì%’"1Ó˜"1ôÓ˜"1IÚ"1`IÚ"1©þ"1Ì©þ"10#1ì0#1ã~#1L ã~#1¥$1¬ ¥$1}ò$1p }ò$1·Þ%1t ·Þ%1{€&1T {€&1KR'1˜KR'1ód'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡: b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1“Ó€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡: b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1“Ó€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡: b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1“Ó €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1!A  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84«4«>?I‘I‘ST^w^whêhêr[s[} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=×=×?J@JB½B½D0E0G£G£IJL‰L‰NüNüPoQoSâSâUUVUXÈXÈZ;[;]®]®_!`!b”b”degzgziíiík`l`nÓnÓpFqFs¹s¹u,v,xŸxŸz{}…}…øø�k‚k„Þ„Þ†O‡O‰Á‰Á‹3Œ3ޥޥ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1w}!1�w}!1¾!1 ¾!1ÅÂ!1ÅÂ!1Iç!1€Iç!1ç'"1üç'"1—,"1—,"1Q"1pQ"1¹‘"1칑"1i–"1ôi–"1…Ú"1`…Ú"1åþ"1Ìåþ"10#1ì0#1ã~#1L ã~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1½€&1T ½€&1�R'1˜�R'15e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxË— b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1(h€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxË— b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1(h€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxË— b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1(h €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Zz  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝSSÉÉ)?*?4µ4µ>)?)I�I�ST^…^…hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��yyïïe e"Û"Û$Q%Q'Ç'Ç)=*=,³,³.)/)1Ÿ1Ÿ346‹6‹89;w;w=í=í?c@cBÙBÙDOEOGÅGÅI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒuŒuŽéŽé�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1é}!1�é}!1‡¾!1 ‡¾!1/Å!1/Å!1µç!1€µç!1S("1üS("1/"1/"1‡Q"1p‡Q"1%’"1ì%’"1Ó˜"1ôÓ˜"1gÚ"1`gÚ"1Çþ"1ÌÇþ"10#1ì0#1ã~#1L ã~#1“$1¬ “$1kò$1p kò$1¥Þ%1t ¥Þ%1€&1T €&1åQ'1˜åQ'1�d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l : b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1“Ó€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l : b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1“Ó€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l : b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1“Ó €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1<\  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛNNÁÁ)4*44¥4¥>?I‡I‡SøSø]i^ihÚhÚrIsI} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹þþqqääW W"Ê"Ê$=%='°'°)#*#,–,–. / 1|1|3ï3ï5b6b8Õ8Õ:H;H=¹=¹?*@*B›B›D E G}G}IîIîK_L_NÐNÐPAQAS²S²U#V#X”X”Z[]v]v_ç_çaXbXdÉdÉf:g:i«i«kln�n�pþpþrosouàuàwQxQzÂzÂ|3}3¤¤�‚„†„††õ†õˆe‰e‹Õ‹Õ�EŽE�µ�µ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1w}!1�w}!1¾!1 ¾!1§Â!1§Â!1Cç!1€Cç!1á'"1üá'"1y,"1y,"1Q"1pQ"1³‘"1쳑"1K–"1ôK–"1[Ú"1`[Ú"1»þ"1Ì»þ"10#1ì0#1Ý~#1L Ý~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1ù€&1T ù€&1ÉR'1˜ÉR'1qe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw̘ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw̘ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw̘ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations12R  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>-?-I¢I¢ST^Ž^Žhisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbçbçd\e\gÑgÑiFjFl»l»n2o2q§q§stv‘v‘xy{{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1é}!1�é}!1‡¾!1 ‡¾!1SÅ!1SÅ!1»ç!1€»ç!1Y("1üY("1%/"1%/"1�Q"1p�Q"1+’"1ì+’"1÷˜"1ô÷˜"1+Ú"1`+Ú"1‹þ"1Ì‹þ"10#1ì0#1Ý~#1L Ý~#1·$1¬ ·$1ßð$1p ßð$1½À%1t ½À%1À&1T À&1¹P'1˜¹P'1ac'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªC b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1½ý€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªC b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1½ý€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªC b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1½ý €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1&"  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚOOÄÄ)9*94®4®>"?"I–I–ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""šš„„ùùn n"ã"ã$X%X'Í'Í)B*B,·,·.,/,1¡1¡346‹6‹89;u;u=ê=ê?_@_BÔBÔDIEIG¾G¾I2J2L¦L¦NOQŽQŽSTVvVvXêXêZ^[^]Ò]Ò_F`FbÄbÄd8e8g¬g¬i j l”l”noq|q|sðsðudvdxØxØzL{L}À}À4€4‚¨‚¨„…‡�‡�‰ŠŒzŒzŽïŽï�d‘d“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1q}!1�q}!1¾!1 ¾!1¡Â!1¡Â!1Cç!1€Cç!1á'"1üá'"1s,"1s,"1Q"1pQ"1³‘"1쳑"1E–"1ôE–"1Ú"1`Ú"1ßþ"1Ìßþ"10#1ì0#1Ý~#1L Ý~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÈ” b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÈ” b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÈ” b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1[{  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÅÅ):*:4¯4¯>"?"I•I•ST^^hòhòrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=ß=ß?T@TBÉBÉD>E>G³G³I&J&L™L™N O QQSòSòUeVeXØXØZK[K]¾]¾_1`1b¤b¤degŠgŠiýiýkplpnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L+i!1é}!1�é}!1‡¾!1 ‡¾!1SÅ!1SÅ!1»ç!1€»ç!1Y("1üY("1%/"1%/"1�Q"1p�Q"1+’"1ì+’"1÷˜"1ô÷˜"1IÚ"1`IÚ"1©þ"1Ì©þ"10#1ì0#1Ý~#1L Ý~#1«$1¬ «$1Óð$1p Óð$1±À%1t ±À%1{€&1T {€&1qP'1˜qP'1c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©C b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¼ü€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©C b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¼ü€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©C b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¼ü €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1IE  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØMMÂÂ)7*74¬4¬>?I’I’S T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?M@MBÂBÂD7E7G¬G¬IJL’L’NOQxQxSëSëU^V^XÑXÑZD[D]·]·_*`*b¤b¤degˆgˆiúiúklllnÞnÞpPqPsÂsÂu4v4x¦x¦z{}Š}Šüü�n‚n„à„à†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1k}!1�k}!1 ¾!1  ¾!1¹Â!1¹Â!1=ç!1€=ç!1Û'"1üÛ'"1‹,"1‹,"1Q"1pQ"1­‘"1ì­‘"1]–"1ô]–"1CÚ"1`CÚ"1£þ"1Ô£þ"10#1ô0#1Ý~#1T Ý~#1½$1´ ½$1åð$1x åð$1ÃÀ%1| ÃÀ%1Õ€&1\ Õ€&1ËP'1 ËP'1sc'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÉ• b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1$d€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÉ• b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1$d€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÉ• b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1$d €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1¹‘>  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee ß ßUUÍÍ)C*C4·4·>+?+IŸIŸST^“^“hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯%%›› ‡ ‡"ý"ý$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@ƒ@ƒB÷B÷DkEkGßGßISJSLÇLÇN;O;Q¯Q¯S#T#V—V—X Y [[]ó]ó_g`gbçbçd[e[gÏgÏiCjCl·l·n+o+qŸqŸstv‡v‡xûxûzo{o}ã}ãW€W‚˂˄?…?‡±‡±‰$Š$Œ—Œ—Ž � ‘}‘}“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1ï}!1�ï}!1�¾!1 �¾!1)Å!1)Å!1Áç!1€Áç!1_("1ü_("1û."1û."1“Q"1p“Q"11’"1ì1’"1͘"1ô͘"1«Ø"1`«Ø"1 ý"1Ì ý"10#1ì0#1Ý~#1L Ý~#1Ã$1¬ Ã$1ëð$1p ëð$1ÉÀ%1t ÉÀ%1ó€&1T ó€&1éP'1˜éP'1‘c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³L b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1“Ó€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³L b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1“Ó€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³L b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1“Ó €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1­©  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ Õ ÕLLÍÍ)D*D4»4»>2?2I©I©S T ^ž^žhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÕÕLLÃà : : ± ±((ŸŸ��{{òòi i"à"à$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÑLÑNHOHQ¿Q¿S6T6V­V­X$Y$[›[›]^`‰`‰bcewewgîgîiejelÜlÜnZoZqÌqÌs>t>v°v°x"y"{”{”}~€x€x‚ê‚ê„\…\‡·Ή@Š@Œ²Œ²Ž$�$‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1k}!1�k}!1 ¾!1  ¾!1¹Â!1¹Â!17ç!1€7ç!1Õ'"1üÕ'"1‹,"1‹,"1 Q"1p Q"1§‘"1ì§‘"1]–"1ô]–"1gÚ"1`gÚ"1Çþ"1ÔÇþ"10#1ô0#1Ý~#1T Ý~#1¥$1´ ¥$1Íð$1x Íð$1«À%1| «À%1c€&1\ c€&1YP'1 YP'1c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÅ‘ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1%e€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÅ‘ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1%e€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1>rÅ‘ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1%e €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1¹‘@d  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛOOÇÇ);*;4­4­>?I‘I‘ST^�^�hóhórcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››ƒƒ÷÷k k"ß"ß$W%W'Ë'Ë)?*?,³,³.'/'1›1›346ƒ6ƒ8÷8÷:k;k=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b—b—d e g{g{iíiík_l_nÑnÑpCqCsµsµu'v'x™x™z { }}}}ïï�a‚a„Ó„Ó†C‡C‰´‰´‹%Œ%Ž–Ž–�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1ï}!1�ï}!1�¾!1 �¾!1)Å!1)Å!1Áç!1€Áç!1_("1ü_("1û."1û."1“Q"1p“Q"11’"1ì1’"1͘"1ô͘"1«Ø"1`«Ø"1 ý"1Ì ý"10#1ì0#1Ý~#1L Ý~#1±$1¬ ±$1Ùð$1p Ùð$1·À%1t ·À%1™€&1T ™€&1�P'1˜�P'17c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²L b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1“Ó€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²L b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1“Ó€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²L b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1“Ó €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1­©  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ Ó ÓHHÅÅ):*:4¯4¯>$?$I™I™ST^Œ^Œhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÓÓHH½½ 2 2 § §‘‘{{ððeeÚÚO O"Ä"Ä$A%A'¶'¶)+*+, , ./1Š1Š3ÿ3ÿ5t6t8é8é:^;^=Ó=Ó?H@HB½B½D2E2G§G§IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚noqrqrsäsäuVvVxÈxÈz:{:}¬}¬€‚�‚�„…‡t‡t‰æ‰æ‹XŒXŽÊŽÊ�<‘<“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„%i!1A}!1A}!1ß½!1|ß½!1CÃ!1„CÃ!1yÆ!1ðyÆ!1Ð!1|Ð!19Ó!1H9Ó!1 ç!1Ô ç!1«'"1P«'"1-"1X-"1K0"1ÄK0"1Õ9"1PÕ9"1 ="1 ="1ßP"1¨ßP"1}‘"1$ }‘"1ç–"1, ç–"1š"1˜ š"1§£"1$ §£"1ݦ"1ð ݦ"1Ô"1| Ô"1Ú"1XÚ"1ßþ"1Ôßþ"1!,#1ô!,#1I:#1ÐI:#1o#1\o#1õ�#1€õ�#1ó˜#1¬ó˜#1·$18·$1�ò$1ü�ò$1ÉÞ%1ÉÞ%1Cì%1àCì%1À&1lÀ&1“R'1°“R'1;e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P§s b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¯ï bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P§s b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¯ï  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P§s  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¯ï bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¬Ì¯Ïf† bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1M�Úi©]�ø8  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dMÍšçg´4�NΛèh·7Û[†°0UÕ… $ ¤ Z Ú ó s  B ‘  ` à - m º ú I‰­íØ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10== × ×LLËË)@*@4µ4µ>+?+IŸIŸST^“^“his}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð==µµ--¥¥   · ·//§§——  ��öök k"à"à$_%_'Ô'Ô)I*I,¾,¾.3/31¨1¨346’6’89;|;|=ñ=ñ?f@fBÛBÛDPEPGÅGÅI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÏbÏdCeCg·g·i+j+lŸlŸnoq‡q‡sûsûuovoxãxãzW{W}Ë}Ë?€?‚³‚³„'…'‡�‡�‰ŠŒ‡Œ‡ŽüŽü�q‘q“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1Ÿ¾!1dŸ¾!1MÅ!1lMÅ!1×Î!1è×Î!1 Ò!1´ Ò!1Óç!1@Óç!1q("1¼q("1/"1Ä/"1©8"1@©8"1ß;"1 ß;"1¥Q"1˜¥Q"1C’"1C’"1ñ˜"1 ñ˜"1{¢"1˜ {¢"1±¥"1d ±¥"1óÒ"1ð óÒ"1‹Ú"1Ì ‹Ú"1ëþ"1H ëþ"1-,#1h-,#1 :#1D :#1Én#1ÐÉn#1Ž#1ôŽ#1™#1 ™#1½$1¬½$1•ò$1p•ò$1ÏÞ%1tÏÞ%1Iì%1TIì%1ù€&1àù€&1ÉR'1$ÉR'1qe'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜Ìd0 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ?¡á¾þ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜Ìd0 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ?¡á¾þ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜Ìd0  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ?¡á¾þ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Šª'Gp� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1?¾þ=}m­¼ü  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d?¿þ~½=|ü;»úz¹9xø7·övƒµ5Ü\tô·7 3 ³ �  ò r ± 1 p ð / o ® î - m ¬ ì ÿ? bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÉÉ)@*@4•4•>.?.I£I£ST^™^™hisƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=á=á?X@XBÏBÏDFEFG½G½IVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbèbèd]e]gÒgÒiGjGl¼l¼n=o=q²q²s't'vœvœxy{†{†}û}ûp€p‚å‚å„Z…Z‡χωDŠDŒ¹Œ¹Ž.�.‘£‘£“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„%i!1;}!1;}!1Ù½!1|Ù½!1CÃ!1„CÃ!1yÆ!1ðyÆ!1Ð!1|Ð!19Ó!1H9Ó!1 ç!1Ô ç!1«'"1P«'"1-"1X-"1K0"1ÄK0"1Õ9"1PÕ9"1 ="1 ="1ßP"1¨ßP"1}‘"1$ }‘"1ç–"1, ç–"1š"1˜ š"1§£"1$ §£"1ݦ"1ð ݦ"1Ô"1| Ô"1=Ú"1X=Ú"1�þ"1Ô�þ"1ß+#1ôß+#1 ;#1Ð ;#1Åo#1\Åo#1£‘#1€£‘#1¡œ#1¬¡œ#1«$18«$1ƒò$1üƒò$1½Þ%1½Þ%17ì%1à7ì%1�€&1l�€&1QR'1°QR'1ùd'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¤p b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¯ï bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¤p b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¯ï  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¤p  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¯ï bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Šª¯Ï)I bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1~¾<|ø8ò2 ´ô  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d~þ|üzúxøvötôròpðlìh è d ä / ¯  „ ` à Ù Y \ Ü ® . X Ø TÔPÐB‚NŽ LÈ„Ä bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10<< Õ ÕJJÇÇ)<*<4±4±>&?&I™I™ST^‰^‰hûhûroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð<<²²((žž   ­ ­##™™……úúooääY Y"Î"Î$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD%?%I˜I˜S T ^Š^Šhýhýrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=½=½?2@2B§B§DEG‘G‘I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnònòpeqesØsØuKvKx¾x¾z1{1}¤}¤€‚ЂЄý„ý†p‡p‰ã‰ã‹VŒVŽÉŽÉ�<‘<“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1k}!1èk}!1 ¾!1d ¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1=ç!1@=ç!1Û'"1¼Û'"11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1Q"1˜Q"1­‘"1­‘"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1Ú"1Ì Ú"1ßþ"1H ßþ"1!,#1h!,#1+:#1D+:#1çn#1Ðçn#1é�#1ôé�#1ç˜#1 ç˜#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1ç€&1àç€&1·R'1$·R'1_e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÊ– b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 0p?È bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÊ– b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 0p?È bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?sÊ–  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 0p?È  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ©É¯Ïb‚ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1D„ÈNŽZšÔ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dDĈÌL�TÔ˜Ü\  fæäd,¬»;òr’ ¸8 ~ þ i é D Ä  Š Ð P ” Ô  X ž Þ $dÊ  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Ö ÖMMÄÄ);*;4²4²> ? I¢I¢ST^•^•hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÖÖMMÄÄ ; ; ² ²))  ŽŽ||óój j"á"á$X%X'Ï'Ï)F*F,½,½.4/41«1«3"4"6™6™89;‡;‡=þ=þ?u@uBìBìDcEcGÚGÚI1J1L¨L¨NOQ–Q–S T V¦V¦XY[”[”] ^ `‚`‚bùbùdpepgçgçi^j^lÕlÕnQoQqÃqÃs5t5v§v§xy{‹{‹}ý}ýo€o‚á‚á„S…S‡Ňʼn7Š7Œ©Œ©Ž�‘�‘�“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1û}!1èû}!1™¾!1d™¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1Íç!1@Íç!1k("1¼k("11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1ŸQ"1˜ŸQ"1=’"1=’"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1IÚ"1Ì IÚ"1©þ"1H ©þ"1ë+#1hë+#1ß:#1Dß:#1›o#1Лo#1Ë�#1ôË�#1É›#1 É›#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1Ï€&1àÏ€&1ŸR'1$ŸR'1Ge'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�ÁY% b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 JŠ‘ÑÉ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�ÁY% b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 JŠ‘ÑÉ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�ÁY%  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 JŠ‘ÑÉ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¯Ï‹«-M bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1q±"bl¬ÕˆÈ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dqñbâSÓDÄ5µ&¦—ˆŽû{éiîn ÄD Ÿ  á a Ô T Ç G º : ­ -  \‘ÑB‚õ5¨è bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff à àVVÌÌ)B*B4¸4¸>,?,I I ST^Š^ŠhÜhÜrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffààZZÔÔ N N È ÈBB¼¼66°°&&œœ ˆ ˆ"þ"þ$t%t'ê'ê)`*`,Ö,Ö.L/L1Â1Â38486®6®8$9$;š;š=>@„@„BøBøDlElGàGàITJTLÈLÈNü>üH”I”S T ^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÔÔII¾¾ 3 3 ¨ ¨’’||ññffÛÛP P"Å"Å$:%:'¯'¯)$*$,™,™./1ƒ1ƒ3ø3ø5m6m8â8â:W;W=Ì=Ì?A@AB¶B¶D+E+G G IôIôKiLiNÞNÞPSQSSÈSÈU`V`XÕXÕZJ[J]¿]¿_4`4b©b©deg“g“ijl}l}n÷n÷piqisÛsÛuMvMx¿x¿z1{1}£}£€‚‡‚‡„ù„ù†k‡k‰݉݋OŒOŽÁŽÁ�3‘3“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1Ÿ¾!1dŸ¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1Íç!1@Íç!1k("1¼k("11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1ŸQ"1˜ŸQ"1=’"1=’"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1mÚ"1Ì mÚ"1Íþ"1H Íþ"1,#1h,#1a:#1Da:#1o#1Ðo#1ÓŽ#1ôÓŽ#1Ñ™#1 Ñ™#1Ÿ$1¬Ÿ$1wò$1pwò$1±Þ%1t±Þ%1+ì%1T+ì%1W€&1àW€&1'R'1$'R'1Ïd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ã]) b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹‘ÑÊ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ã]) b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹‘ÑÊ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ã])  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹‘ÑÊ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¯Ï’²Tt bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1T”è(zº L G  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dTÔ(¨ü|ÐP¤$xøLÌ  òrÄD– h è : º z ú  Œ U Õ Þ ^ 0 ° ° 0 ‹ „ Ä Xªê<|·÷ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜPPÄÄ)8*84¬4¬>?I�I�ST^x^xhÇhÇr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœ„„øøl l"à"à$T%T'È'È)<*<,°,°.$/$1˜1˜3 4 6€6€8ô8ô:h;h=Ü=Ü?N@NBÀBÀD2E2G¤G¤IJLˆLˆNúNúPlQlSÞSÞUPVPXÂXÂZ4[4]¦]¦_`bŽbŽdegrgriäiäkVlVnÈnÈp:q:s¬s¬uvx�x�zßzß|Q}QÃÃ�5‚5„§„§†<‡<‰­‰­‹ŒŽ�Ž��‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1q}!1Ðq}!1¾!1L¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1Cç!1(Cç!1á'"1¤á'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1Q"1€Q"1³‘"1ü³‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1©7#1,©7#1Ž#1˜Ž#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1í€&1Ìí€&1½R'1½R'1ee'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tË— b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T?Ó bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tË— b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T?Ó bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tË—  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T?Ó  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¯Ï¨È`€ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1U•€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ëkÄD� v ö å% €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ý ÝWWÍÍ)C*C4¹4¹>/?/I£I£ST^‹^‹hÿhÿrsss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÝÝWWÑÑ K K Å Å??¹¹33­­''¡¡ • •"#%…%…'û'û)q*q,ç,ç.]/]1Ó1Ó3I4I6¿6¿8595;«;«=!>!@—@—B C EƒEƒGùGùIoJoLãLãNWOWQËQËS?T?V³V³X'Y'[›[›]^`ƒ`ƒb÷b÷dkekgßgßiSjSlÇlÇn;o;q¯q¯s#t#v—v—x y {{}ó}óg€g‚Û‚Û„O…O‡ÇÉ7Š7Œ«Œ«Ž�‘“‘““ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1õ}!1Ðõ}!1“¾!1L“¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Áç!1(Áç!1_("1¤_("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#1»7#1,»7#1[Ž#1˜[Ž#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1#�&1Ì#�&1óR'1óR'1›e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶N b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•ؾþ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶N b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•ؾþ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶N  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•ؾþ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¯ÏŒ¬3S bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1g§€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1•òrÏO ¬ , 7w €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff ß ßXXÑÑ)J*J4Ã4Ã>7@°@°B)C)E¢E¢GHJ”J”L M O‚O‚QùQùSpTpVçVçX^Y^[Õ[Õ]L^L`Ã`Ãb8c8e­e­g"h"j—j—l m ooqòqòsetevØvØxKyK{¾{¾}1~1€¤€¤‚ƒ…ЅЇý‡ý‰oŠoŒáŒáŽS�S‘őœ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1q}!1Ðq}!1¾!1L¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1=ç!1(=ç!1Û'"1¤Û'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1Q"1€Q"1­‘"1ü­‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#1C:#1,C:#1£Ž#1˜£Ž#1Ÿ$1$Ÿ$1wò$1èwò$1±Þ%1ì±Þ%1{€&1Ì{€&1KR'1KR'1ód'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÉ• b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U?Ò bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÉ• b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U?Ò bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÉ•  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U?Ò  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¯Ïˆ¨-M bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1î. €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1O Ï ( ¨  � Ú Z ~¾ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSÇÇ);*;4¯4¯>#?#I•I•ST^y^yhëhër]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$g%g'Û'Û)O*O,Ã,Ã.7/71«1«346“6“89;{;{=ï=ï?c@cB×B×DKEKG¿G¿I3J3L¥L¥NOQ‰Q‰SûSûUmVmXßXßZQ[Q]Ã]Ã_5`5b§b§deg‹g‹iýiýkolonánápSqSsÅsÅu7v7x©x©z{}�}�ÿÿ�q‚q„ã„ã†U‡U‰ljNj9Œ9ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1õ}!1Ðõ}!1“¾!1L“¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Çç!1(Çç!1e("1¤e("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1[:#1,[:#1÷Ž#1˜÷Ž#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1«€&1Ì«€&1{R'1{R'1#e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸R b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V–Ú¿ÿ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸R b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V–Ú¿ÿ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸R  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V–Ú¿ÿ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¯Ï—·^~ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1‰ f æ C à   Ô €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I£I£ST^‡^‡høhørjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbæbædYeYgÌgÌi?j?l²l²n#o#q”q”stvvvvxçxçzX{X}É}É:€:‚«‚«„…‡އމÿ‰ÿ‹pŒpŽáŽá�R‘R“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1w}!1Ðw}!1¾!1L¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Cç!1(Cç!1á'"1¤á'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1Q"1€Q"1³‘"1ü³‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1©7#1,©7#1%Ž#1˜%Ž#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1u€&1Ìu€&1ER'1ER'1íd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au̘ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 %eA�à  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au̘ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 %eA�à  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au̘  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 %eA�à   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 §Ç'G^~ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1P�€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1òrÍM¨( ƒ  @ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ö ÖKKÀÀ)5*54ª4ª>?I�I�ST^v^vhéhér\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††ûûppååZ Z"Ï"Ï$D%D'¹'¹).*.,£,£./1�1�346w6w8ì8ì:a;a=Ö=Ö?I@IB¼B¼D/E/G¢G¢IJLˆLˆNûNûPnQnSáSáUTVTXÇXÇZ:[:]­]­_ ` b“b“degygyiìiìk_l_nÒnÒpEqEs¸s¸u+v+xžxžz{}„}„÷÷�j‚j„݄݆P‡P‰‰‹4Œ4ަަ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ï}!1Ðï}!1�¾!1L�¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1»ç!1(»ç!1Y("1¤Y("1[/"1¬[/"1å8"1(å8"1<"1ô<"1�Q"1€�Q"1+’"1ü+’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1»7#1,»7#1yŽ#1˜yŽ#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%15�&1Ì5�&1S'1S'1­e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«C b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ` ‘Ñî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«C b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ` ‘Ñî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«C  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ` ‘Ñî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¯Ï�­5U bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1b¢€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1œû{ÚZ ¹ 9 R’ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÎÎ)F*F4¾4¾>6?6I¬I¬S"T"^˜^˜his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$n%n'æ'æ)^*^,Ö,Ö.N/N1Æ1Æ3>4>6¶6¶8.9.;¦;¦=>@–@–BCE†E†GþGþIvJvLìLìNbObQØQØSNTNVÄVÄX:Y:[°[°]&^&`œ`œbceˆeˆgþgþitjtlêlên`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1q}!1Ðq}!1¾!1L¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Cç!1(Cç!1á'"1¤á'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1Q"1€Q"1³‘"1ü³‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#1C:#1,C:#1ÁŽ#1˜ÁŽ#1“$1$“$1kò$1èkò$1¥Þ%1ì¥Þ%1€&1Ì€&1åQ'1åQ'1�d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÊ– b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 $d@€Þ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÊ– b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 $d@€Þ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÊ–  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 $d@€Þ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‡§$/O bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ë+ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1^ Þ 9 ¹  ” ï o ›Û €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØKK¾¾)1*14¤4¤>?I†I†S÷S÷]h^hhÙhÙrJsJ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆûûnnááT T"Ç"Ç$:%:'­'­) * ,“,“./1y1y3ì3ì5_6_8Ò8Ò:E;E=¸=¸?)@)BšBšD E G|G|IíIíK^L^NÏNÏP@Q@S±S±U"V"X“X“Z[]u]u_æ_æaWbWdÈdÈf9g9iªiªklnŒnŒpýpýrnsnußußwPxPzÁzÁ|2}2££�‚„…„…†ö†öˆf‰f‹Ö‹Ö�FŽF�¶�¶’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ï}!1Ðï}!1�¾!1L�¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Áç!1(Áç!1_("1¤_("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1“Q"1€“Q"11’"1ü1’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#1[:#1,[:#1�#1˜�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­G b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡‘Ñð0 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­G b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡‘Ñð0 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­G  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡‘Ñð0  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¯Ï˜¸`€ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1A €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1˜  w ÷ V Ö 5 µ ñ1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÆÆ)<*<4²4²>(?(IœIœST^„^„høhørlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$N%N'Ä'Ä):*:,°,°.&/&1œ1œ346ˆ6ˆ8þ8þ:t;t=ê=ê?`@`BÖBÖDLELGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡Œ‡Œ‰ŠŒtŒtŽèŽè�\‘\“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1}}!1Ð}}!1¾!1L¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1Iç!1(Iç!1ç'"1¤ç'"1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1Q"1€Q"1¹‘"1ü¹‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1©7#1,©7#1CŽ#1˜CŽ#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÍ™ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IW—@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÍ™ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IW—@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÍ™  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IW—@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¤ÄFfXx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1K‹€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ùyÖV³3 �  [ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛPPÅÅ):*:4¯4¯>%?%I™I™S T ^�^�hõhõrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››……úúo o"ä"ä$Y%Y'Î'Î)C*C,¸,¸.-/-1¢1¢346Œ6Œ89;v;v=ë=ë?`@`BÕBÕDJEJG¿G¿I5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`Ib½b½d1e1g¥g¥ijl�l�noququséséu]v]xÑxÑzE{E}¹}¹-€-‚¡‚¡„…‡‹‡‹‰ŠŒuŒuŽêŽê�_‘_“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ã}!1Ðã}!1�¾!1L�¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1µç!1(µç!1S("1¤S("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1‡Q"1€‡Q"1%’"1ü%’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#1»7#1,»7#1—Ž#1˜—Ž#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1ÿ€&1Ìÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l 8 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ` ²òâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l 8 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ` ²òâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l 8  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ` ²òâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¯Ï†¦'G bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1]�€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1#£„åe Æ F m­ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ý ÝTTËË)B*B4¹4¹>0?0I¥I¥ST^�^�hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÝÝTTËË B B ¹ ¹00§§••  ƒƒúúq q"è"è$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubêbêd_e_gÔgÔiIjIl¾l¾n3o3q¨q¨stv’v’xy{|{|}ñ}ñf€f‚Û‚Û„P…P‡Ňʼn:Š:Œ¯Œ¯Ž$�$‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1w}!1Ðw}!1¾!1L¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1Iç!1(Iç!1ç'"1¤ç'"1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1Q"1€Q"1¹‘"1ü¹‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1I:#1,I:#1åŽ#1˜åŽ#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvË— b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JV–þ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvË— b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JV–þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvË—  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JV–þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 „¤'G/O bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1è( €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1m í J Ê ' §  „ ¸ø €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84­4­>"?"I•I•ST^y^yhëhër_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?N@NBÃBÃD8E8G­G­I"J"L•L•NOQ{Q{SîSîUaVaXÔXÔZG[G]º]º_-`-bŸbŸdegƒgƒiõiõkglgnÙnÙpKqKs½s½u/v/x¡x¡z{}…}…÷÷�i‚i„Û„Û†O‡O‰‰‹5Œ5ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1é}!1Ðé}!1‡¾!1L‡¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1»ç!1(»ç!1Y("1¤Y("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1�Q"1€�Q"1+’"1ü+’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1[:#1,[:#19�#1˜9�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢< b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡³óä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢< b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡³óä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢<  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡³óä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¯Ï‘±Rr bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1þ> €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1§ ' ˆ  i é J Ê N €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÆÆ);*;4°4°>%?%I˜I˜S T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnænæpYqYsÌsÌu?v?x²x²z%{%}˜}˜ € ‚~‚~„ñ„ñ†d‡d‰׉׋JŒJ޽޽�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1}}!1Ð}}!1¾!1L¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Oç!1(Oç!1í'"1¤í'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1!Q"1€!Q"1¿‘"1ü¿‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1©7#1,©7#1gŽ#1˜gŽ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1í€&1Ìí€&1½R'1½R'1ee'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwΚ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Z?É  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwΚ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Z?É  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwΚ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Z?É   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 —·Mm>^ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1F†€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ß_¾> �  6v €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>/?/I¦I¦ST^�^�hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIWJWLÎLÎNEOEQ¼Q¼S3T3VªVªX!Y![˜[˜]^`†`†býbýdtetgëgëibjblÙlÙnKoKq½q½s/t/v¡v¡xy{…{…}÷}÷i€i‚Û‚Û„M…M‡¿‡¿‰1Š1Œ£Œ£Ž�‘‡‘‡“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1Ý}!1ÐÝ}!1{¾!1L{¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1¯ç!1(¯ç!1M("1¤M("1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1�Q"1€�Q"1’"1ü’"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1µ7#1,µ7#1‘Ž#1˜‘Ž#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1á€&1Ìá€&1±R'1±R'1Ye'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•-ù b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I‰È•Õ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•-ù b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I‰È•Õ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•-ù  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I‰È•Õ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 {›5a� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1T”€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1šû{Ü\ ½ = d¤ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ä äZZÐÐ)F*F4¼4¼>0?0I¤I¤ST^Œ^Œhistst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjää^^ØØ R R Ì ÌFFÀÀ::´´**   Œ Œ"#%x%x'î'î)d*d,Ú,Ú.P/P1Æ1Æ3<4<6²6²8(9(;ž;ž=>@ˆ@ˆBüBüDpEpGäGäIXJXLÌLÌN@O@Q´Q´S(T(VœVœXY[„[„]ø]ø_l`lbàbàdTeTgÈgÈi#?#I˜I˜S T ^^hñhñrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccØØMM 7 7 ¬ ¬!!––  €€õõjjßßT T"É"É$>%>'³'³)(*(,�,�./1‡1‡3ü3ü5q6q8æ8æ:[;[=Ð=Ð?E@EBºBºD/E/G¤G¤IJL�L�NOQzQzSïSïUdVdXÙXÙZN[N]Ã]Ã_8`8b­b­d"e"g—g—i j l�l�nónópeqes×s×uIvIx»x»z-{-}Ÿ}Ÿ€‚ƒ‚ƒ„õ„õ†g‡g‰ىًKŒK޽޽�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1ã}!1Ðã}!1�¾!1L�¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1µç!1(µç!1S("1¤S("1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1‡Q"1€‡Q"1%’"1ü%’"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1U:#1,U:#13�#1˜3�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1i€&1Ìi€&19R'19R'1ád'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 HˆÇ–Ö bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 HˆÇ–Ö bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 HˆÇ–Ö  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ŠªNnDd bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1õ5 €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ž   ÿ ` à A Á E €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àTTÈÈ)<*<4°4°>"?"I”I”ST^x^xhêhêr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  ˆˆüüp p"ä"ä$X%X'Ì'Ì)@*@,´,´.(/(1œ1œ346„6„8ø8ø:l;l=à=à?R@RBÄBÄD6E6G¨G¨IJLŒLŒNþNþPpQpSâSâUTVTXÆXÆZ8[8]ª]ª_`bŽbŽdegrgriäiäkVlVnÈnÈp:q:s¬s¬uvx�x�z{}t}tææ�X‚X„ʄʆ<‡<‰­‰­‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1ƒ}!1Ѓ}!1!¾!1L!¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Uç!1(Uç!1ó'"1¤ó'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1'Q"1€'Q"1Å‘"1üÅ‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1©7#1,©7#1‹Ž#1˜‹Ž#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1�&1Ì�&1ÕR'1ÕR'1}e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÏ› b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @@€Í  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÏ› b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @@€Í  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÏ›  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @@€Í   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ŠªHh$D bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Cƒ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ‰ëkÍM ¯ / [› €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á á[[ÑÑ)G*G4½4½>3?3I§I§ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±++¥¥ ™ ™"#%‰%‰'ÿ'ÿ)u*u,ë,ë.a/a1×1×3M4M6Ã6Ã8999;¯;¯=%>%@›@›BCE‡E‡GýGýIsJsLçLçN[O[QÏQÏSCTCV·V·X+Y+[Ÿ[Ÿ]^`‡`‡bûbûdoeogãgãiWjWlËlËn?o?q³q³s't'v›v›xy{ƒ{ƒ}÷}÷k€k‚߂߄S…S‡LJlj;Š;Œ¯Œ¯Ž#�#‘—‘—“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1×}!1Ð×}!1u¾!1Lu¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1©ç!1(©ç!1G("1¤G("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1{Q"1€{Q"1’"1ü’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1©7#1,©7#1gŽ#1˜gŽ#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%15�&1Ì5�&1S'1S'1­e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ"î b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 LŒË ©é bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ"î b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 LŒË ©é bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ"î  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 LŒË ©é  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p�#^~ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Cƒ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ý}Ü\»; š  3s €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ã ã\\ÕÕ)N*N4Ç4Ç>@?@I·I·S,T,^Ÿ^Ÿhis…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjãã\\ÕÕ N N Ç Ç@@¹¹22««$$�� � �"#%�%�'ú'ú)s*s,ì,ì.e/e1Þ1Þ3W4W6Ð6Ð8I9I;Â;Â=;>;@´@´B-C-E¦E¦GHJ˜J˜LMO†O†QýQýStTtVëVëXbYb[Ù[Ù]P^P`Ç`Çb'?'I™I™S T ^}^}hïhïrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ��  "÷"÷$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1Ý}!1ÐÝ}!1{¾!1L{¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1©ç!1(©ç!1G("1¤G("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1{Q"1€{Q"1’"1ü’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1I:#1,I:#1�#1˜�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ&ò b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹Ê ©é bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ&ò b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹Ê ©é bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ&ò  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹Ê ©é  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ƒ£6V]} bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1â" €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1y ù X Ø 7 ·  – Ò €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>2?2I§I§ST^‹^‹hühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbêbêd]e]gÐgÐiCjCl¶l¶n'o'q˜q˜s t vzvzxëxëz\{\}Í}Í>€>‚¯‚¯„ … ‡’‡’‰ŠŒtŒtŽåŽå�V‘V“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1‰}!1Љ}!1'¾!1L'¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1[ç!1([ç!1ù'"1¤ù'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1-Q"1€-Q"1Ë‘"1üË‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1=Ú"1´ =Ú"1�þ"10 �þ"1ß+#1Pß+#1»7#1,»7#1�#1˜�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyМ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O?Ï bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyМ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O?Ï bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyМ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O?Ï  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 }�¯Ï * bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1NŽ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18¸!¡ Š ó s Î €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94®4®>!?!I”I”ST^z^zhíhír`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?M@MBÀBÀD3E3G¦G¦IJLŒLŒNÿNÿPrQrSåSåUXVXXËXËZ>[>]±]±_$`$b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†T‡T‰ƉƋ8Œ8ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1Ñ}!1ÐÑ}!1o¾!1Lo¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1£ç!1(£ç!1A("1¤A("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1uQ"1€uQ"1’"1ü’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1»7#1,»7#1åŽ#1˜åŽ#1á$1$á$1¹ò$1è¹ò$1óÞ%1ìóÞ%1G�&1ÌG�&1S'1S'1¿e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kã b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O�Πà bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kã b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O�Πà bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kã  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O�Î à  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 e…ñWw bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1NŽ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18¸� ‚ ç g ®î €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âZZÒÒ)J*J4Â4Â>:?:I°I°S&T&^œ^œhisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢’’  ‚ ‚"ú"ú$r%r'ê'ê)b*b,Ú,Ú.R/R1Ê1Ê3B4B6º6º8292;ª;ª=">"@š@šBCEŠEŠGHJzJzLðLðNfOfQÜQÜSRTRVÈVÈX>Y>[´[´]*^*` ` bceŒeŒghjxjxlîlîndodqÚqÚsPtPvÆvÆx?IŠIŠSûSû]l^lhÝhÝrNsN} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒÿÿrrååX X"Ë"Ë$>%>'±'±)$*$,—,—. / 1}1}3ð3ð5c6c8Ö8Ö:I;I=¼=¼?-@-BžBžDEG€G€IñIñKbLbNÓNÓPDQDSµSµU&V&X—X—Z[]y]y_ê_êa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1×}!1Ð×}!1u¾!1Lu¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1£ç!1(£ç!1A("1¤A("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1uQ"1€uQ"1’"1ü’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1[:#1,[:#1‡�#1˜‡�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ç b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 NŽÍ ¡á bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ç b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 NŽÍ ¡á bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ç  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 NŽÍ ¡á  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 |œ$D^~ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ó3 €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ì L ± 1 –  { û S“ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞTTÊÊ)@*@4¶4¶>,?,I I ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽzzððf f"Ü"Ü$R%R'È'È)>*>,´,´.*/*1 1 346Œ6Œ89;x;x=î=î?d@dBÚBÚDPEPGÆGÆI8?8I¬I¬S T ^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðxxððhhàà X X Ð ÐHHÀÀ88°°%%šš „ „"ù"ù$n%n'ã'ã)X*X,Í,Í.B/B1·1·3,4,6¡6¡89;‹;‹=>@u@uBêBêD_E_GÔGÔIHJHL¼L¼N0O0Q¤Q¤STVŒVŒXY[t[t]è]è_\`\bÐbÐdDeDg¸g¸i,j,l l noqˆqˆsüsüupvpxäxäzX{X}Ì}Ì@€@‚´‚´„(…(‡œ‡œ‰ŠŒ†Œ†ŽûŽû�p‘p“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ë}!1�Ë}!1i¾!1 i¾!1eÅ!1eÅ!1�ç!1€�ç!1;("1ü;("17/"17/"1oQ"1poQ"1 ’"1ì ’"1 ™"1ô ™"1‹Ú"1`‹Ú"1ëþ"1Ìëþ"1/0#1ì/0#1û~#1L û~#1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1�&1T �&1íR'1˜íR'1•e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s × b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1µõ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s × b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1µõ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s × b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1µõ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ii  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss ê êaaØØ)O*O4Æ4Æ>;?;I°I°S%T%^š^šhis„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssêêaaØØ O O Æ Æ==´´++¢¢�� ~ ~"õ"õ$l%l'ã'ã)Z*Z,Ñ,Ñ.H/H1¿1¿36466­6­8$9$;›;›=>@‰@‰BCEwEwGîGîIcJcLØLØNMOMQÂQÂS7T7V¬V¬X!Y![–[–] ^ `€`€bõbõdjejgßgßiTjTlÉlÉn>o>q³q³s(t(v�v�xy{‡{‡}ü}üq€q‚æ‚æ„[…[‡ЇЉEŠEŒºŒºŽ/�/‘¤‘¤“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1×}!1�×}!1u¾!1u¾!1³Â!1$³Â!1£ç!1�£ç!1A("1A("1…,"1$…,"1uQ"1�uQ"1’"1’"1W–"1$W–"1íØ"1�íØ"1Mý"1üMý"1/0#1 /0#1û~#1| û~#1Ï$1Ü Ï$1§ò$1  §ò$1áÞ%1¤ áÞ%1Ÿ€&1„ Ÿ€&1oR'1ÈoR'1e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations12fU‰Þªä b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations12fU‰Þªä b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations12fU‰Þªä b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¯Ï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ì ìaaÖÖ)K*K4À4À>3?3I¦I¦ST^Š^Šhühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvììbbØØ N N Ä Ä::°°&&œœ††ûûp p"å"å$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ì=ì?a@aBÖBÖDKEKGÀGÀI3J3L¦L¦NOQŒQŒSÿSÿUrVrXåXåZX[X]Ë]Ë_>`>b°b°d"e"g”g”ijlxlxnênêp\q\sÎsÎu@v@x²x²z${$}–}–€‚z‚z„ì„ì†^‡^‰щыDŒDŽ·Ž·�*‘*“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ñ}!1�Ñ}!1o¾!1 o¾!1eÅ!1eÅ!1�ç!1€�ç!1;("1ü;("17/"17/"1oQ"1poQ"1 ’"1ì ’"1 ™"1ô ™"1‘Ú"1`‘Ú"1ñþ"1Ìñþ"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1·€&1T ·€&1‡R'1˜‡R'1/e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s Ù b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1µõ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s Ù b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1µõ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1?s Ù b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1µõ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Rr  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ê ê``ÕÕ)J*J4¿4¿>2?2I¥I¥ST^‹^‹hþhþrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttêê``ÖÖ L L  Â88®®$$šš††üür r"è"è$]%]'Ò'Ò)G*G,¼,¼.1/11¦1¦346�6�89;z;z=ï=ï?d@dBÙBÙDNENGÃGÃI6J6L©L©NOQ�Q�STVuVuXèXèZ[[[]Î]Î_A`Ab´b´d'e'gšgši j l€l€nónópfqfsÙsÙuLvLx¿x¿z2{2}¥}¥€‚‹‚‹„þ„þ†q‡q‰ä‰ä‹WŒWŽÊŽÊ�=‘=“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1›}!1�›}!19¾!1 9¾!1¹Â!1¹Â!1mç!1€mç!1 ("1ü ("1‹,"1‹,"1?Q"1p?Q"1Ý‘"1ìÝ‘"1]–"1ô]–"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1 �&1T  �&1ÛR'1˜ÛR'1ƒe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆß« b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1O€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆß« b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1O€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆß« b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1O €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ë ëbbÙÙ)P*P4Ç4Ç>>?>IµIµS,T,^ž^žhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttëëbbÙÙ P P Ç Ç>>µµ,,££‘‘  "ö"ö$m%m'ä'ä)[*[,Ò,Ò.I/I1À1À37476®6®8%9%;œ;œ=>@Š@ŠBCExExGïGïIfJfLÝLÝNTOTQËQËSBTBV¹V¹X0Y0[§[§]^`•`•b c eƒeƒgúgúiqjqlèlènZoZqÌqÌs>t>v°v°x"y"{”{”}~€x€x‚ê‚ê„\…\‡·Ή@Š@Œ²Œ²Ž$�$‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ñ}!1�Ñ}!1o¾!1 o¾!1MÅ!1MÅ!1£ç!1€£ç!1A("1üA("1/"1/"1uQ"1puQ"1’"1ì’"1ñ˜"1ôñ˜"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1ó€&1T ó€&1ÃR'1˜ÃR'1ke'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|à b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 à€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|à b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 à€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|à b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 à €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¯Ï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss í íccÙÙ)O*O4Ã4Ã>7?7I«I«ST^“^“hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssííggáá [ [ Õ ÕOOÉÉCC½½33©© • •" # %�%�'÷'÷)m*m,ã,ã.Y/Y1Ï1Ï3E4E6»6»8191;§;§=>@�@�BCEwEwGëGëI_J_LÓLÓNGOGQ»Q»S/T/V£V£XY[‹[‹]ÿ]ÿ_s`sbçbçd[e[gÏgÏiCjCl·l·n+o+qŸqŸstv‡v‡xûxûzo{o}ã}ãW€W‚˂˄?…?‡±‡±‰$Š$Œ—Œ—Ž � ‘}‘}“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1›}!1�›}!19¾!1 9¾!1¹Â!1¹Â!1gç!1€gç!1("1ü("1‹,"1‹,"19Q"1p9Q"1ב"1ìב"1]–"1ô]–"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1±€&1T ±€&1�R'1˜�R'1)e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†Û§ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†Û§ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†Û§ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¯Ï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ç ç\\ÑÑ)F*F4»4»>0?0I¥I¥ST^Œ^Œhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrçç\\ÑÑ F F » »00¥¥��yyîîc c"Ø"Ø$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=ß=ß?T@TBÉBÉD>E>G³G³I(J(L�L�NOQ‡Q‡SüSüUqVqXæXæZ[[[]Ð]Ð_E`Ebºbºd/e/g¤g¤ijlŽlŽnoqrqrsäsäuVvVxÈxÈz:{:}¬}¬€‚�‚�„…‡t‡t‰æ‰æ‹XŒXŽÊŽÊ�<‘<“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ñ}!1�Ñ}!1o¾!1 o¾!1MÅ!1MÅ!1£ç!1€£ç!1A("1üA("1/"1/"1uQ"1puQ"1’"1ì’"1ñ˜"1ôñ˜"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1{€&1T {€&1KR'1˜KR'1ód'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|â b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¡á€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|â b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¡á€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|â b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¡á €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¯Ï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss ë ë__ÓÓ)G*G4¹4¹>+?+I�I�ST^�^�hóhórcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssëëccÛÛ S S Ë ËCC»»33««““ { {"ï"ï$c%c'×'×)K*K,¿,¿.3/31§1§346�6�89;w;w=é=é?[@[BÍBÍD?E?G±G±I#J#L•L•NOQyQySëSëU]V]XÏXÏZA[A]³]³_%`%b—b—d e g{g{iíiík_l_nÑnÑpCqCsµsµu'v'x™x™z { }}}}ïï�a‚a„Ó„Ó†C‡C‰´‰´‹%Œ%Ž–Ž–�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1•}!1�•}!13¾!1 3¾!1³Â!1³Â!1gç!1€gç!1("1ü("1…,"1…,"19Q"1p9Q"1ב"1ìב"1W–"1ôW–"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ܨ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ܨ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ܨ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¯Ï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt î îhhÞÞ)T*T4Ê4Ê>>?>I²I²S&T&^š^šhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttîîhhââ \ \ Ö ÖPPÊÊDD¾¾88²²,, ¦ ¦" # %–%–' ( *‚*‚,ø,ø.n/n1ä1ä3Z4Z6Ð6Ð8F9F;¼;¼=2>2@¨@¨BCE”E”G H J~J~LòLòNfOfQÚQÚSNTNVÂVÂX6Y6[ª[ª]^`’`’bcezezgîgîibjblÖlÖnJoJq¾q¾s2t2v¦v¦xy{Ž{Ž}~€v€v‚ê‚ê„^…^‡Ò‡Ò‰FŠFŒºŒºŽ.�.‘¢‘¢“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1×}!1�×}!1u¾!1 u¾!1eÅ!1eÅ!1©ç!1€©ç!1G("1üG("17/"17/"1{Q"1p{Q"1’"1ì’"1 ™"1ô ™"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1;�&1T ;�&1 S'1˜ S'1³e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…é b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1½ý€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…é b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1½ý€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…é b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1½ý €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¯Ï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo è èaaÚÚ)S*S4Ì4Ì>C?CIºIºS/T/^¢^¢his†s†} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooèèaaÚÚ S S Ì ÌEE¾¾77°°))¢¢ ” ”" # %†%†'ÿ'ÿ)x*x,ñ,ñ.j/j1ã1ã3\4\6Õ6Õ8N9N;Ç;Ç=@>@@¹@¹B2C2E«E«G$H$J›J›LMO‰O‰QRTwTwVîVîXeYe[Ü[Ü]S^S`Ê`Êb?c?e´e´g)h)jžjžlmo†o†qùqùsltlvßvßxRyR{Å{Å}8~8€«€«‚ƒ…‘…‘‡ˆŠtŠtŒæŒæŽX�X‘ʑʓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1�}!1��}!1-¾!1 -¾!1³Â!1³Â!1aç!1€aç!1ÿ'"1üÿ'"1…,"1…,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1W–"1ôW–"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OƒØ¤ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OƒØ¤ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OƒØ¤ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¯Ï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ê êbbÖÖ)J*J4¾4¾>0?0I¢I¢ST^†^†høhørjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrêêbbÚÚ R R Ê ÊBBºº22ªª""šš Š Š"#%v%v'ê'ê)^*^,Ò,Ò.F/F1º1º3.4.6¢6¢89;Š;Š=þ=þ?r@rBæBæDZEZGÎGÎI@J@L²L²N$O$Q–Q–STVzVzXìXìZ^[^]Ð]Ð_B`Bb´b´d&e&g˜g˜i j l|l|nînîp`q`sÒsÒuDvDx¶x¶z({(}š}š € ‚~‚~„ð„ð†b‡b‰Ô‰Ô‹FŒFޏޏ�*‘*“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1×}!1�×}!1u¾!1 u¾!1eÅ!1eÅ!1©ç!1€©ç!1G("1üG("17/"17/"1{Q"1p{Q"1’"1ì’"1 ™"1ô ™"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ë b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¼ü€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ë b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¼ü€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ë b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¼ü €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¯Ï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo æ æ]]ÔÔ)K*K4Â4Â>7?7I¬I¬ST^�^�hisqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooææ]]ÔÔ K K  Â99°°''žžŒŒ z z"ñ"ñ$h%h'ß'ß)V*V,Í,Í.D/D1»1»32426©6©8 9 ;—;—=>@…@…BüBüDsEsGêGêI_J_LÔLÔNIOIQ¾Q¾S3T3V¨V¨XY[’[’]^`|`|bïbïdbebgÕgÕiHjHl»l»n,o,q�q�stvvxðxðza{a}Ò}ÒC€C‚´‚´„%…%‡•‡•‰ŠŒwŒwŽèŽè�Y‘Y“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1�}!1��}!1-¾!1 -¾!1ËÂ!1ËÂ!1aç!1€aç!1ÿ'"1üÿ'"1�,"1�,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1o–"1ôo–"1gÚ"1`gÚ"1Çþ"1ÌÇþ"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚Ù¥ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1&f€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚Ù¥ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1&f€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚Ù¥ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1&f €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations12R  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ã ãXXÍÍ)B*B4µ4µ>(?(I›I›ST^�^�hôhôrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““}}òòg g"Ü"Ü$Q%Q'Æ'Æ);*;,°,°.%/%1š1š346„6„8ù8ù:n;n=á=á?T@TBÇBÇD:E:G­G­I J L“L“NOQyQySìSìU_V_XÒXÒZE[E]¸]¸_+`+bžbždeg„g„i÷i÷kjljnÝnÝpPqPsÃsÃu6v6x©x©z{}�}�€‚u‚u„è„è†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Ý}!1�Ý}!1{¾!1 {¾!15Å!15Å!1©ç!1€©ç!1G("1üG("1/"1/"1{Q"1p{Q"1’"1ì’"1Ù˜"1ôÙ˜"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1A�&1T A�&1S'1˜S'1¹e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ&ò b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‘Ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ&ò b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‘Ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ&ò b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‘Ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¯Ï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk ã ã[[ÓÓ)K*K4Ã4Ã>9?9I¯I¯S%T%^›^›his‡s‡} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkãã[[ÓÓ K K à Ã;;³³++££““  ƒ ƒ"û"û$s%s'ë'ë)c*c,Û,Û.S/S1Ë1Ë3C4C6»6»8393;«;«=#>#@›@›BCE‹E‹GHJyJyLïLïNeOeQÛQÛSQTQVÇVÇX=Y=[³[³])^)`Ÿ`Ÿbce‹e‹ghjwjwlílíncocqÙqÙsOtOvÅvÅx;y;{±{±}'~'€�€�‚ƒ…‰…‰‡ÿ‡ÿ‰uŠuŒëŒëŽa�a‘בד b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1‰}!1�‰}!1'¾!1 '¾!1ËÂ!1ËÂ!1[ç!1€[ç!1ù'"1üù'"1�,"1�,"1-Q"1p-Q"1Ë‘"1ìË‘"1o–"1ôo–"1sÚ"1`sÚ"1Óþ"1ÌÓþ"1#0#1ì#0#1ï~#1L ï~#1¥$1¬ ¥$1}ò$1p }ò$1·Þ%1t ·Þ%1'€&1T '€&1÷Q'1˜÷Q'1Ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Õ¡ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1'g€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Õ¡ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1'g€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Õ¡ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1'g €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1=]  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ã ãVVÉÉ)<*<4­4­>?I�I�ST^q^qhâhârQsQ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““yyìì_ _"Ò"Ò$E%E'¸'¸)+*+,ž,ž./1„1„3÷3÷5j6j8Ý8Ý:P;P=Á=Á?2@2B£B£DEG…G…IöIöKgLgNØNØPIQISºSºU+V+XœXœZ [ ]~]~_ï_ïa`b`dÑdÑfBgBi³i³k$l$n•n•pqswswuèuèwYxYzÊzÊ|;};¬¬�‚„ބކý†ýˆm‰m‹Ý‹Ý�MŽM�½�½’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Ý}!1�Ý}!1{¾!1 {¾!15Å!15Å!1¯ç!1€¯ç!1M("1üM("1/"1/"1�Q"1p�Q"1’"1ì’"1Ù˜"1ôÙ˜"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1Ï€&1T Ï€&1ŸR'1˜ŸR'1Ge'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ(ô b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‘Ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ(ô b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‘Ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ(ô b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‘Ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¯Ï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk á áWWÍÍ)C*C4¹4¹>-?-I¡I¡ST^‰^‰hýhýrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkááWWÍÍ C C ¹ ¹//¥¥‘‘}}óói i"ß"ß$U%U'Ë'Ë)A*A,·,·.-/-1£1£346�6�89;{;{=ñ=ñ?g@gBÝBÝDSESGÉGÉI=J=L±L±N%O%Q™Q™S T V�V�XõXõZi[i]Ý]Ý_Q`QbÅbÅd9e9g­g­i!j!l•l•n o q}q}sñsñuevexÙxÙzM{M}Á}Á5€5‚©‚©„…‡‘‡‘‰ŠŒyŒyŽíŽí�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1‰}!1�‰}!1'¾!1 '¾!1³Â!1³Â!1Uç!1€Uç!1ó'"1üó'"1…,"1…,"1'Q"1p'Q"1Å‘"1ìÅ‘"1W–"1ôW–"1gÚ"1`gÚ"1Çþ"1ÌÇþ"10#1ì0#1é~#1L é~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÖ¢ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÖ¢ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÖ¢ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations17W  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ä äYYÎÎ)C*C4¸4¸>,?,I I ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllää\\ÔÔ L L Ä Ä<<´´,,¤¤ŽŽ x x"í"í$b%b'×'×)L*L,Á,Á.6/61«1«3 4 6•6•8 9 ;;=ô=ô?i@iBÞBÞDSESGÈGÈI/?/I¤I¤ST^Ž^Žhisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbébéd^e^gÓgÓiHjHl½l½n2o2q§q§stv‘v‘xy{{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ƒ}!1�ƒ}!1!¾!1 !¾!1³Â!1³Â!1Uç!1€Uç!1ó'"1üó'"1…,"1…,"1'Q"1p'Q"1Å‘"1ìÅ‘"1W–"1ôW–"1yÚ"1`yÚ"1Ùþ"1ÌÙþ"10#1ì0#1é~#1L é~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Òž b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Òž b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Òž b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Jj  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj à àUUÊÊ)?*?4´4´>'?'IšIšS T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjààVVÌÌ B B ¸ ¸..¤¤��zzïïd d"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¤b¤degˆgˆiúiúklllnÞnÞpPqPsÂsÂu4v4x¦x¦z{}Š}Šüü�n‚n„à„à†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ã}!1�ã}!1�¾!1 �¾!1YÅ!1YÅ!1µç!1€µç!1S("1üS("1+/"1+/"1‡Q"1p‡Q"1%’"1ì%’"1ý˜"1ôý˜"1áØ"1`áØ"1Aý"1ÌAý"10#1ì0#1é~#1L é~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¹ù€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¹ù€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¹ù €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¯Ï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞTTÉÉ)>*>4³4³>&?&I™I™S T ^^hòhòrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽzzððf f"Ü"Ü$Q%Q'Æ'Æ);*;,°,°.%/%1š1š346„6„8ù8ù:n;n=ã=ã?X@XBÍBÍDBEBG·G·I*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b¨b¨degŽgŽijltltnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1ƒ}!1�ƒ}!1!¾!1 !¾!1¿Â!1¿Â!1Oç!1€Oç!1í'"1üí'"1‘,"1‘,"1!Q"1p!Q"1¿‘"1ì¿‘"1c–"1ôc–"1gÚ"1`gÚ"1Çþ"1ÌÇþ"10#1ì0#1ã~#1L ã~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1í€&1T í€&1½R'1˜½R'1ee'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|ÓŸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1#c€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|ÓŸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1#c€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|ÓŸ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1#c €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1<\  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>2?2I©I©S T ^’^’hisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÑLÑNHOHQ¿Q¿S6T6V­V­X$Y$[›[›]^`‰`‰bcewewgîgîiejelÜlÜnNoNqÀqÀs2t2v¤v¤xy{ˆ{ˆ}ú}úl€l‚Þ‚Þ„P…P‡‡‰4Š4Œ¦Œ¦Ž�‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1ã}!1�ã}!1�¾!1 �¾!1/Å!1/Å!1µç!1€µç!1S("1üS("1/"1/"1‡Q"1p‡Q"1%’"1ì%’"1Ó˜"1ôÓ˜"1[Ú"1`[Ú"1»þ"1Ì»þ"10#1ì0#1ã~#1L ã~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1Õ€&1T Õ€&1¥R'1˜¥R'1Me'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l 8 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‘Ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l 8 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‘Ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l 8 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‘Ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1.N  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á áWWÍÍ)C*C4·4·>+?+IŸIŸST^‡^‡hûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±''�� ‰ ‰"ÿ"ÿ$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@ƒ@ƒB÷B÷DkEkGßGßISJSLÇLÇN;O;Q¯Q¯S#T#V—V—X Y [[]ó]ó_g`gbÛbÛdOeOgÃgÃi7j7l«l«noq“q“stv{v{xïxïzc{c}×}×K€K‚¿‚¿„3…3‡¥‡¥‰ŠŒ‹Œ‹ŽþŽþ�q‘q“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1}}!1�}}!1¾!1 ¾!1¿Â!1¿Â!1Oç!1€Oç!1í'"1üí'"1‘,"1‘,"1!Q"1p!Q"1¿‘"1ì¿‘"1c–"1ôc–"1Ú"1`Ú"1ßþ"1Ìßþ"10#1ì0#1ã~#1L ã~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÏ› b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1$d€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÏ› b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1$d€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÏ› b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1$d €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Uu  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Û ÛPPÅÅ):*:4¯4¯>$?$I™I™ST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÛÛPPÅÅ : : ¯ ¯$$™™ƒƒøømmââW W"Ì"Ì$A%A'¶'¶)+*+, , ./1Š1Š3ÿ3ÿ5t6t8é8é:^;^=Ó=Ó?H@HB½B½D2E2G§G§IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1é}!1�é}!1‡¾!1 ‡¾!1/Å!1/Å!1µç!1€µç!1S("1üS("1/"1/"1‡Q"1p‡Q"1%’"1ì%’"1Ó˜"1ôÓ˜"1sÚ"1`sÚ"1Óþ"1ÌÓþ"10#1ì0#1ã~#1L ã~#1¥$1¬ ¥$1}ò$1p }ò$1·Þ%1t ·Þ%1]€&1T ]€&1-R'1˜-R'1Õd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l : b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‘Ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l : b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‘Ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l : b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‘Ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ii  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßSSÇÇ);*;4­4­>?I‘I‘ST^u^uhçhçrWsW} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ‡‡ûûo o"ã"ã$W%W'Ë'Ë)?*?,³,³.'/'1›1›346ƒ6ƒ8÷8÷:k;k=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b‹b‹dýdýfogoiáiákSlSnÅnÅp7q7s©s©uvx�x�zÿzÿ|q}qãã�U‚U„DŽdž7‡7‰¨‰¨‹ŒŽŠŽŠ�û�û’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1w}!1�w}!1¾!1 ¾!1§Â!1§Â!1Iç!1€Iç!1ç'"1üç'"1y,"1y,"1Q"1pQ"1¹‘"1칑"1K–"1ôK–"1[Ú"1`[Ú"1»þ"1Ì»þ"10#1ì0#1ã~#1L ã~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1ù€&1T ù€&1ÉR'1˜ÉR'1qe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyМ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyМ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyМ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations15U  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh â â\\ÒÒ)H*H4¾4¾>2?2I¦I¦ST^�^�hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²,,¦¦  š š"#%Š%Š'(*v*v,ì,ì.b/b1Ø1Ø3N4N6Ä6Ä8:9:;°;°=&>&@œ@œBCEˆEˆGþGþIrJrLæLæNZOZQÎQÎSBTBV¶V¶X*Y*[ž[ž]^`†`†búbúdnengâgâiVjVlÊlÊn@o@q´q´s(t(vœvœxy{„{„}ø}øl€l‚à‚à„T…T‡ȇȉ<Š<Œ°Œ°Ž$�$‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1é}!1�é}!1‡¾!1 ‡¾!1MÅ!1MÅ!1»ç!1€»ç!1Y("1üY("1/"1/"1�Q"1p�Q"1+’"1ì+’"1ñ˜"1ôñ˜"17Ú"1`7Ú"1—þ"1Ì—þ"10#1ì0#1ã~#1L ã~#1Ï$1¬ Ï$1÷ð$1p ÷ð$1ÕÀ%1t ÕÀ%1)�&1T )�&1Q'1˜Q'1Çc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©A  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1·÷€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©A  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1·÷€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©A  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1·÷ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations13/  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ü ÜUUÎÎ)G*G4À4À>7?7I®I®S#T#^ ^ his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÜÜUUÎÎ G G À À99²²++¤¤–– ˆ ˆ"#%z%z'ó'ó)l*l,å,å.^/^1×1×3P4P6É6É8B9B;»;»=4>4@­@­B&C&EŸEŸGHJ�J�LMO}O}QôQôSkTkVâVâXYYY[Ð[Ð]G^G`¾`¾b3c3e¨e¨ghj’j’lmo„o„q÷q÷sjtjvÝvÝxPyP{Ã{Ã}6~6€©€©‚ƒ…�…�‡ˆŠrŠrŒäŒäŽV�V‘ȑȓ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1w}!1�w}!1¾!1 ¾!1§Â!1§Â!1Cç!1€Cç!1á'"1üá'"1y,"1y,"1Q"1pQ"1³‘"1쳑"1K–"1ôK–"1…Ú"1`…Ú"1åþ"1Ìåþ"10#1ì0#1ã~#1L ã~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw̘ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw̘ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cw̘ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1^~  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÊÊ)>*>4²4²>$?$I–I–ST^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$j%j'Þ'Þ)R*R,Æ,Æ.:/:1®1®3"4"6–6–8 9 ;~;~=ò=ò?f@fBÚBÚDNENGÂGÂI4J4L¦L¦NOQŠQŠSüSüUnVnXàXàZR[R]Ä]Ä_6`6b¨b¨degŒgŒiþiþkplpnænæpXqXsÊsÊuŒ>ްް�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1ï}!1�ï}!1�¾!1 �¾!1MÅ!1MÅ!1»ç!1€»ç!1Y("1üY("1/"1/"1�Q"1p�Q"1+’"1ì+’"1ñ˜"1ôñ˜"1[Ú"1`[Ú"1»þ"1Ì»þ"10#1ì0#1ã~#1L ã~#1·$1¬ ·$1ßð$1p ßð$1½À%1t ½À%1·€&1T ·€&1­P'1˜­P'1Uc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©C b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¶ö€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©C b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¶ö€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©C b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¶ö €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1VR  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>+?+I I ST^Œ^Œhýhýrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞISJSLÈLÈN=O=Q²Q²S'T'VœVœXY[†[†]û]û_p`pbãbãdVeVgÉgÉi?I‘I‘ST^ƒ^ƒhöhörgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=×=×?J@JB½B½D0E0G£G£IJL‰L‰NüNüPoQoSâSâUUVUXÈXÈZ;[;]®]®_!`!b b deg†g†iùiùklllnßnßpRqRsÅsÅu8v8x«x«z{}‘}‘€‚w‚w„ê„ê†[‡[‰͉͋?Œ?ޱޱ�#‘#“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1ï}!1�ï}!1�¾!1 �¾!1#Å!1#Å!1Áç!1€Áç!1_("1ü_("1õ."1õ."1“Q"1p“Q"11’"1ì1’"1ǘ"1ôǘ"1±Ø"1`±Ø"1ý"1Ìý"10#1ì0#1Ý~#1L Ý~#1Õ$1¬ Õ$1ýð$1p ýð$1ÛÀ%1t ÛÀ%1;�&1T ;�&11Q'1˜1Q'1Ùc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²J b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‘Ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²J b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‘Ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²J b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‘Ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¯«  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ × ×OOÑÑ)I*I4Á4Á>7?7I­I­S#T#^¥^¥his‘s‘} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__××OOÇÇ ? ? · ·//§§——‡‡ÿÿw w"ï"ï$q%q'é'é)a*a,Ù,Ù.Q/Q1É1É3A4A6¹6¹8191;©;©=!>!@™@™BCE‰E‰GHJwJwLíLíNcOcQÙQÙSOTOVÅVÅX;Y;[±[±]'^'`�`�bce‰e‰gÿgÿiujulëlënmomqãqãsYtYvÏvÏxEyE{»{»}1~1€§€§‚ƒ…“…“‡ ˆ ŠŠŒõŒõŽk�k‘á‘á“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1q}!1�q}!1¾!1 ¾!1³Â!1³Â!1=ç!1€=ç!1Û'"1üÛ'"1…,"1…,"1Q"1pQ"1­‘"1ì­‘"1W–"1ôW–"1[Ú"1`[Ú"1»þ"1Ô»þ"10#1ô0#1Ý~#1T Ý~#1™$1´ ™$1Áð$1x Áð$1ŸÀ%1| ŸÀ%1€&1\ €&1P'1 P'1¹b'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÉ• b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1!a€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÉ• b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1!a€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1@tÉ• b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1!a €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1»“2V  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×NNÁÁ)4*44¥4¥>?I‡I‡ST^u^uhæhærUsU} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡þþqqääW W"Ê"Ê$=%='°'°)#*#,–,–. / 1|1|3ï3ï5b6b8Õ8Õ:H;H=¹=¹?*@*B›B›D E G}G}IîIîK_L_NÐNÐPAQAS²S²U#V#X”X”Z[]v]v_ç_çadbddÕdÕfFgFi·i·k(l(n™n™p q s{s{uìuìw]x]zÎzÎ|?}?°°�!‚!„’„’†‡‰q‰q‹á‹á�QŽQ�Á�Á’ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1ï}!1�ï}!1�¾!1 �¾!1#Å!1#Å!1Áç!1€Áç!1_("1ü_("1õ."1õ."1“Q"1p“Q"11’"1ì1’"1ǘ"1ôǘ"1±Ø"1`±Ø"1ý"1Ìý"10#1ì0#1Ý~#1L Ý~#1½$1¬ ½$1åð$1p åð$1ÃÀ%1t ÃÀ%1À&1T À&1¹P'1˜¹P'1ac'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²L b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‘Ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²L b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‘Ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²L b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‘Ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¯«  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Õ ÕKKÉÉ)?*?4µ4µ>)?)I�I�ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……ûûqqçç] ]"Ó"Ó$Q%Q'Ç'Ç)=*=,³,³.)/)1Ÿ1Ÿ346‹6‹89;w;w=í=í?c@cBÙBÙDOEOGÅGÅI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noq…q…sùsùumvmxáxázU{U}É}É=€=‚±‚±„%…%‡™‡™‰ Š Œ�Œ�ŽõŽõ�i‘i“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„%i!1A}!1A}!1ß½!1|ß½!1CÃ!1„CÃ!1yÆ!1ðyÆ!1Ð!1|Ð!19Ó!1H9Ó!1ç!1Ôç!1±'"1P±'"1-"1X-"1K0"1ÄK0"1Õ9"1PÕ9"1 ="1 ="1åP"1¨åP"1ƒ‘"1$ ƒ‘"1ç–"1, ç–"1š"1˜ š"1§£"1$ §£"1ݦ"1ð ݦ"1Ô"1| Ô"1aÚ"1XaÚ"1Áþ"1ÔÁþ"1,#1ô,#1y:#1Ðy:#15o#1\5o#1ÓŽ#1€ÓŽ#1Ñ™#1¬Ñ™#1½$18½$1•ò$1ü•ò$1ÏÞ%1ÏÞ%1Iì%1àIì%1Û€&1lÛ€&1«R'1°«R'1Se'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P©u b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1±ñ bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P©u b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1±ñ  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P©u  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1±ñ bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �­±ÑJj bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations17wX˜ð0ŠÊ$d  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d7·XØ0°–ˆõuà`ÔT¸8�hè@ÀšôtÎN ¨ ( ‚  \ Ü 6 ¶  � è ()i€ÀZ´ô bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10== Û ÛQQÑÑ)G*G4½4½>1?1I¥I¥S%T%^™^™h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð==··11«« % % à Ã==··11««!!——  ƒ ƒ"ù"ù$y%y'ï'ï)e*e,Û,Û.Q/Q1Ç1Ç3=4=6³6³8)9);Ÿ;Ÿ=>@‰@‰BýBýDqEqGåGåIYJYLÍLÍNAOAQµQµS)T)V�V�XY[…[…]ù]ù_m`mbíbídaeagÕgÕiIjIl½l½n1o1q¥q¥stv�v�xy{u{u}é}é]€]‚тфE…E‡¹‡¹‰,Š,ŒŸŒŸŽ�‘…‘…“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1¥¾!1d¥¾!1SÅ!1lSÅ!1ÝÎ!1èÝÎ!1Ò!1´Ò!1Ùç!1@Ùç!1w("1¼w("1%/"1Ä%/"1¯8"1@¯8"1å;"1 å;"1«Q"1˜«Q"1I’"1I’"1÷˜"1 ÷˜"1�¢"1˜ �¢"1·¥"1d ·¥"1ùÒ"1ð ùÒ"1OÚ"1Ì OÚ"1¯þ"1H ¯þ"1ñ+#1hñ+#1Ó:#1DÓ:#1�o#1Ð�o#1¡�#1ô¡�#1Ÿ›#1 Ÿ›#1«$1¬«$1ƒò$1pƒò$1½Þ%1t½Þ%17ì%1T7ì%1{€&1à{€&1KR'1$KR'1ód'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎh4 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 A��Ý bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎh4 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 A��Ý bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎh4  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 A��Ý  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ±Ñ„¤8X bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations19yo¯^Ï€À  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d9¹oï’^ÞëkMÍÄD<¼+«š ‰øxéiÚZ Ë K ¼ < ­ - ž  � Ë €o¯^Ï€À bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×NNÃÃ)*4­4­> ? I“I“ST^…^…høhørksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1l1l3á3á5V6V8Ë8Ë:@;@=Ù=Ù?L@LB¿B¿D2E2G¥G¥IJL‹L‹NþNþPqQqSäSäUWVWXÊXÊZ=[=]°]°_#`#b¢b¢degˆgˆiûiûknlnnánápTqTsÇsÇu:v:x­x­z { }“}“€‚y‚y„ì„ì†_‡_‰щыCŒC޵޵�'‘'“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„%i!1A}!1A}!1ß½!1|ß½!1CÃ!1„CÃ!1yÆ!1ðyÆ!1Ð!1|Ð!19Ó!1H9Ó!1 ç!1Ô ç!1«'"1P«'"1-"1X-"1K0"1ÄK0"1Õ9"1PÕ9"1 ="1 ="1ßP"1¨ßP"1}‘"1$ }‘"1ç–"1, ç–"1š"1˜ š"1§£"1$ §£"1ݦ"1ð ݦ"1Ô"1| Ô"1mÚ"1XmÚ"1Íþ"1ÔÍþ"1,#1ô,#13;#1Ð3;#1ïo#1\ïo#1�’#1€�’#1�#1¬�#1¥$18¥$1}ò$1ü}ò$1·Þ%1·Þ%11ì%1à1ì%1c€&1lc€&13R'1°3R'1Ûd'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¦r b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1±ñ bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¦r b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1±ñ  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¦r  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1±ñ bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 £Ã±ÑVv bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1‰ÉÖR’Yà  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d‰ ’›¤$³3­-’¶6qñPп?ÈHÏO Ö V Ý ] ä d ë k ò rùyÈ€ IÒ™Ù`  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10<< Ù ÙMMÉÉ)=*=4±4±>#?#I•I•ST^…^…h÷h÷risi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð<<´´,,¤¤   ¹ ¹11©©!!™™  ��õõi i"Ý"Ý$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=á=á?S@SBÅBÅD7E7G©G©IJL�L�NÿNÿPqQqSãSãUUVUXÇXÇZ9[9]«]«_`b›b›d e ggiñiñkclcnÕnÕpGqGs¹s¹u+v+x�x�z{}�}�óó�e‚e„ׄ׆I‡I‰º‰º‹+Œ+ŽœŽœ� ‘ “# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1 ~!1è ~!1«¾!1d«¾!1SÅ!1lSÅ!1ÝÎ!1èÝÎ!1Ò!1´Ò!1Ùç!1@Ùç!1w("1¼w("1%/"1Ä%/"1¯8"1@¯8"1å;"1 å;"1«Q"1˜«Q"1I’"1I’"1÷˜"1 ÷˜"1�¢"1˜ �¢"1·¥"1d ·¥"1ùÒ"1ð ùÒ"1sÚ"1Ì sÚ"1Óþ"1H Óþ"1,#1h,#1;#1D;#1×o#1Ð×o#1÷‘#1ô÷‘#1õœ#1 õœ#1™$1¬™$1qò$1pqò$1«Þ%1t«Þ%1%ì%1T%ì%1€&1à€&1ëQ'1$ëQ'1“d'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐl8 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 A�žÞ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐl8 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 A�žÞ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐl8  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 A�žÞ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ±Ñ¤Ä\| bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1‚ÂÒD„DÄ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d‚„†ˆ�Š vöŒ OÏ(¨Ž�� �  �  �  �  � �d¤�’ÒT”TÔ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×NNÁÁ)*4§4§>?I‰I‰ST^w^whèhèrYsY} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡þþqqääW W"Ê"Ê$=%='°'°)#*#,–,–. / 1W1W3Ê3Ê5=6=8°8°:#;#=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éafbfd×d×fHgHi¹i¹k*l*n›n›p q s}s}uîuîw_x_zÐzÐ|A}A²²�#‚#„”„”†‡‰u‰u‹å‹å�UŽU�Å�Å’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1q}!1èq}!1¾!1d¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1Cç!1@Cç!1á'"1¼á'"11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1Q"1˜Q"1³‘"1³‘"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1mÚ"1Ì mÚ"1Íþ"1H Íþ"1,#1h,#1[:#1D[:#1o#1Ðo#1%Ž#1ô%Ž#1#™#1 #™#1É$1¬É$1¡ò$1p¡ò$1ÛÞ%1tÛÞ%1Uì%1TUì%1#�&1à#�&1óR'1$óR'1›e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuΚ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ,lA�È bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuΚ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ,lA�È bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuΚ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ,lA�È  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �°±ÑPp bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations14tO�Þo¯@  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d4´OÏž‘ímîn¼<ËK‹ ZÚ)©øxÉIškë< ¼  � Þ ^ ¯ / € O � Þ Ho¯@ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ú ÚSSÌÌ)E*E4¾4¾>?I®I®S#T#^ ^ his†s†} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÚÚSSÌÌ E E ¾ ¾77°°))¢¢””  † †"ÿ"ÿ$x%x'ñ'ñ)j*j,ã,ã.\/\1Õ1Õ3N4N6Ç6Ç8@9@;¹;¹=2>2@«@«B$C$E�E�GHJkJkLâLâNYOYQÐQÐSGTGVâVâXYYY[Ð[Ð]G^G`¾`¾b3c3e¨e¨ghj’j’lmo„o„q÷q÷sjtjvÝvÝxPyP{Ã{Ã}6~6€©€©‚ƒ…�…�‡ˆŠtŠtŒæŒæŽX�X‘ʑʓ  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1Ÿ¾!1dŸ¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1Íç!1@Íç!1k("1¼k("11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1ŸQ"1˜ŸQ"1=’"1=’"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1[Ú"1Ì [Ú"1»þ"1H »þ"1ý+#1hý+#1µ:#1Dµ:#1qo#1Ðqo#1ó�#1ôó�#1ñš#1 ñš#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1½€&1ཀ&1�R'1$�R'15e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ã]) b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Hˆ“ÓÉ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ã]) b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Hˆ“ÓÉ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ã])  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Hˆ“ÓÉ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ±Ñ‰©Bb bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations14tf¦ L´ô\œ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d4´fæ�LÌêj2²ÅE˜þ~ädÊJ°0˜€ h è P Ð 8 ¸     ˆ ð pä $Ö|¼$dÌ  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜQQÆÆ);*;4°4°>&?&IšIšST^„^„hÔhÔrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœ††ûûp p"å"å$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ì=ì?a@aBÖBÖDKEKGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`JbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔz${$}˜}˜ € ‚€‚€„ô„ô†އމŠŒxŒxŽíŽí�b‘b“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1q}!1èq}!1¾!1d¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1=ç!1@=ç!1Û'"1¼Û'"11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1Q"1˜Q"1­‘"1­‘"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1mÚ"1Ì mÚ"1Íþ"1H Íþ"1,#1h,#1U:#1DU:#1o#1Ðo#1Ž#1ôŽ#1™#1 ™#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1±€&1à±€&1�R'1$�R'1)e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au̘ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 +kA�Ç bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au̘ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 +kA�Ç bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Au̘  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 +kA�Ç  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¢Â±ÑTt bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1P�à Ñn®ü<  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dPÐ  ðpÀ@�¨(`à…0°bâ€AÁÎNœjê8 ¸  † Ô T ¢ " p ð @ € Ð ^ž¥åì, bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>?I I ST^Œ^Œhýhýroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜI.J.L£L£NOQ�Q�STVœVœXY[†[†]û]û_p`pbãbãdVeVgÉgÉi!?!I”I”S T ^|^|hÉhÉrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?M@MBÂBÂD7E7G¬G¬I!J!L”L”NOQzQzSíSíU`V`XÓXÓZF[F]¹]¹_,`,b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤zñzñ|c}cÕÕ�G‚G„¹„¹†R‡R‰ʼnŋ8Œ8ޫޫ�‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1w}!1Ðw}!1¾!1L¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1Cç!1(Cç!1á'"1¤á'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1Q"1€Q"1³‘"1ü³‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1�2#1,�2#1ùŒ#1˜ùŒ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1)�&1Ì)�&1ùR'1ùR'1¡e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÏ› b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TA�Ó bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÏ› b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TA�Ó bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÏ›  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TA�Ó  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ±Ñ“³Vv bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1'g€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1'§‚Ý]¸8×  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ù ÙQQÉÉ)A*A4¹4¹>1?1I§I§ST^“^“h i ss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A ¹ ¹11©©!!™™‰‰ y y"ñ"ñ$i%i'á'á)Y*Y,Ñ,Ñ.I/I1Á1Á39496±6±8)9);¡;¡=>@‘@‘B C E�E�GùGùIqJqLçLçN]O]QÓQÓSITIV¿V¿X5Y5[«[«]!^!`—`—b c eƒeƒgùgùiojolålån[o[qÑqÑsGtGv½v½x3y3{©{©}~€•€•‚ ƒ …�…�‡÷‡÷‰mŠmŒãŒãŽY�Y‘ϑϓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1û}!1Ðû}!1™¾!1L™¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Çç!1(Çç!1e("1¤e("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1mÚ"1´ mÚ"1Íþ"10 Íþ"1,#1P,#1�2#1,�2#1�#1˜�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1ç€&1Ìç€&1·R'1·R'1_e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸R b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“Ø·÷ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸R b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“Ø·÷ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸R  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“Ø·÷  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ±ÑŽ®Ll bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1+k€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1+«ˆåeÂBû ; €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>0?0I§I§ST^�^�histst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIXJXLÏLÏNFOFQ½Q½S4T4V«V«X"Y"[™[™]^`‡`‡bþbþdueugìgìicjclÚlÚnLoLq¾q¾s0t0v¢v¢xy{†{†}ø}øj€j‚܂܄N…N‡À‡À‰2Š2Œ¤Œ¤Ž�‘ˆ‘ˆ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1q}!1Ðq}!1¾!1L¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1Cç!1(Cç!1á'"1¤á'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1Q"1€Q"1³‘"1ü³‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#15#1,5#1•�#1˜•�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1·€&1Ì·€&1‡R'1‡R'1/e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÍ™ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UA�Ò bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÍ™ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UA�Ò bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÍ™  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UA�Ò  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¡Á±ÑRr bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1“nîIÉ$¤r² €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×MMÃÃ)9*94¯4¯>%?%I™I™S T ^�^�hõhõrisi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$K%K'Á'Á)7*7,­,­.#/#1™1™346…6…8û8û:q;q=ç=ç?]@]BÓBÓDIEIG¿G¿I5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`Ib½b½d1e1g¥g¥ijl�l�noququséséu]v]xÑxÑzE{E}¹}¹-€-‚¡‚¡„…‡‰‡‰‰ý‰ý‹qŒqŽåŽå�Y‘Y“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1û}!1Ðû}!1™¾!1L™¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Íç!1(Íç!1k("1¤k("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1ŸQ"1€ŸQ"1=’"1ü=’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#1!5#1,!5#1¹�#1˜¹�#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºV" b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T”Ú¸ø bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºV" b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T”Ú¸ø bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºV"  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T”Ú¸ø  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ±ÑªÊd„ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1È€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ÿ|üYÙ6¶˜Ø €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb × ×LLÁÁ)6*64«4«>"?"I—I—S T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbb××LLÁÁ 6 6 « «  ••  ôôiiÞÞS S"È"È$=%='²'²)'*',œ,œ./1†1†3û3û5p6p8å8å:Z;Z=Ï=Ï?D@DB¹B¹D.E.G£G£IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1}}!1Ð}}!1¾!1L¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Iç!1(Iç!1ç'"1¤ç'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1Q"1€Q"1¹‘"1ü¹‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1u2#1,u2#1ÏŒ#1˜ÏŒ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1ó€&1Ìó€&1ÃR'1ÃR'1ke'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwМ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 !aCƒà  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwМ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 !aCƒà  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwМ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 !aCƒà   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ±Ñ–¶\| bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Z€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1šósÌL¥%ª ê  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>+?+I I ST^Š^Šhÿhÿrtst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜISJSLÈLÈN=O=Q²Q²S'T'VœVœXY[†[†]û]û_p`pbåbådZeZgÏgÏiDjDl¹l¹n.o.q£q£stv�v�xy{w{w}ì}ìa€a‚Ö‚Ö„K…K‡À‡À‰5Š5ŒªŒªŽ�‘”‘”“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ï}!1Ðï}!1�¾!1L�¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Áç!1(Áç!1_("1¤_("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1“Q"1€“Q"11’"1ü1’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1{2#1,{2#1;�#1˜;�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1ù€&1Ìù€&1ÉR'1ÉR'1qe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­G b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ^ž�Ïî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­G b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ^ž�Ïî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­G  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ^ž�Ïî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ±Ñ“³Vv bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1"b€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1"¢�à`¿?R €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh â â\\ÒÒ)H*H4¾4¾>4?4I¨I¨ST^�^�hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²,,¦¦  š š"#%Š%Š'(*v*v,ì,ì.b/b1Ø1Ø3N4N6Ä6Ä8:9:;°;°=&>&@œ@œBCEˆEˆGþGþItJtLèLèN\O\QÐQÐSDTDV¸V¸X,Y,[ [ ]^`ˆ`ˆbübüdpepgägäiXjXlÌlÌn@o@q´q´s(t(vœvœxy{„{„}ø}øl€l‚à‚à„T…T‡ȇȉ<Š<Œ°Œ°Ž$�$‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1w}!1Ðw}!1¾!1L¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Iç!1(Iç!1ç'"1¤ç'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1Q"1€Q"1¹‘"1ü¹‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#15#1,5#1k�#1˜k�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwΚ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  `B‚Þ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwΚ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  `B‚Þ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwΚ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  `B‚Þ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ±Ñ ÀPp bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1³ó€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1~þW×0° ‰Cƒ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÃÃ)8*84­4­>"?"I•I•ST^{^{hîhîrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD*>4²4²>&?&I˜I˜S T ^|^|hîhîr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$j%j'Þ'Þ)R*R,Æ,Æ.:/:1®1®3"4"6–6–8 9 ;~;~=ò=ò?f@fBÚBÚDNENGÂGÂI6J6L¨L¨NOQŒQŒSþSþUpVpXâXâZT[T]Æ]Æ_8`8bªbªdegŽgŽijlrlrnänäpVqVsÈsÈu:v:x¬x¬z{}�}�€‚t‚t„æ„æ†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1}}!1Ð}}!1¾!1L¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1Oç!1(Oç!1í'"1¤í'"1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1!Q"1€!Q"1¿‘"1ü¿‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1o2#1,o2#1«Œ#1˜«Œ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Õ€&1ÌÕ€&1¥R'1¥R'1Me'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÑ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IY™@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÑ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IY™@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÑ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IY™@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ™¹;[b‚ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 M€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 �äd»;’} ½  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee ß ßUUËË)A*A4·4·>+?+IŸIŸST^‡^‡hûhûroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯%%›› ‡ ‡"ý"ý$s%s'é'é)_*_,Õ,Õ.K/K1Á1Á37476­6­8#9#;™;™=>@ƒ@ƒB÷B÷DkEkGßGßISJSLÇLÇN;O;Q¯Q¯S#T#V—V—X Y [[]ó]ó_g`gbÛbÛdOeOgÃgÃi7j7l«l«noq“q“stv{v{xïxïzc{c}×}×K€K‚¿‚¿„3…3‡§‡§‰ŠŒ�Œ�Ž�‘s‘s“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1é}!1Ðé}!1‡¾!1L‡¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1»ç!1(»ç!1Y("1¤Y("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1�Q"1€�Q"1+’"1ü+’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1{2#1,{2#1S�#1˜S�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢< b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ^ž¬ìâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢< b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ^ž¬ìâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢<  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ^ž¬ìâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ”´SsXx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Y€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1™úzÛ[¼<)i €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛPPÅÅ):*:4¯4¯>"?"I•I•ST^{^{hîhîrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹uuêê_ _"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?N@NBÁBÁD4E4G§G§IJL�L�NOQsQsSæSæUYVYXÌXÌZ?[?]²]²_%`%b˜b˜d e g~g~iñiñkdldn×n×pJqJs½s½u0v0x£x£z{}‰}‰üü�o‚o„â„â†U‡U‰ljNj9Œ9ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1}}!1Ð}}!1¾!1L¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1Oç!1(Oç!1í'"1¤í'"1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1!Q"1€!Q"1¿‘"1ü¿‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#15#1,5#1A�#1˜A�#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1c€&1Ìc€&13R'13R'1Ûd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÏ› b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JX˜þ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÏ› b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JX˜þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÏ›  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JX˜þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ÿ¿<Nn bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¤ä€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ié@À—înT €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝQQÅÅ)9*94­4­>?I‘I‘ST^u^uhçhçrYsY} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��……ùùm m"á"á$U%U'É'É)=*=,±,±.%/%1™1™3 4 6�6�8õ8õ:i;i=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b‹b‹dýdýfogoiáiákSlSnÅnÅp7q7s©s©uvx�x�zÿzÿ|q}qãã�U‚U„DŽdž9‡9‰ª‰ª‹ŒŽŒŽŒ�ý�ý’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ï}!1Ðï}!1�¾!1L�¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Áç!1(Áç!1_("1¤_("1[/"1¬[/"1å8"1(å8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#15#1,5#1õ�#1˜õ�#1™$1$™$1qò$1èqò$1«Þ%1ì«Þ%1€&1Ì€&1ëQ'1ëQ'1“d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤@  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _Ÿ­íä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤@  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _Ÿ­íä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤@   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _Ÿ­íä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¤Ä8XXx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ºú€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�~þ_ß@ÀÊ  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛNNÁÁ)4*44§4§>?I‰I‰SúSú]k^khÜhÜrMsM} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹þþqqääW W"Ê"Ê$=%='°'°)#*#,–,–. / 1|1|3ï3ï5b6b8Õ8Õ:H;H=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éaZbZdËdËf;?;I²I²S'T'^š^šh i s€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÞÞWWÐÐ I I  Â;;´´--¦¦˜˜ Š Š"#%|%|'õ'õ)n*n,ç,ç.`/`1Ù1Ù3R4R6Ë6Ë8D9D;½;½=6>6@¯@¯B(C(E¡E¡GHJ“J“L M O�O�QøQøSoToVæVæX]Y][Ô[Ô]K^K`Â`Âb7c7e¬e¬g!h!j–j–l m o~o~qñqñsdtdv×v×xJyJ{½{½}0~0€£€£‚ƒ…‰…‰‡ü‡ü‰nŠnŒàŒàŽR�R‘đē b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1ã}!1Ðã}!1�¾!1L�¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1µç!1(µç!1S("1¤S("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1‡Q"1€‡Q"1%’"1ü%’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1u2#1,u2#1M�#1˜M�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G‡È“Ó bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G‡È“Ó bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—1ý  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G‡È“Ó  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �­Jjf† bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1P€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�ñqÒR³3 ` €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àUUÊÊ)?*?4´4´>*?*IžIžST^†^†húhúrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  ŠŠÿÿt t"é"é$^%^'Ó'Ó)H*H,½,½.2/21§1§346‘6‘89;{;{=ð=ð?e@eBÚBÚDOEOGÄGÄI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡�‡�‰ŠŒzŒzŽïŽï�d‘d“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1ƒ}!1Ѓ}!1!¾!1L!¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Uç!1(Uç!1ó'"1¤ó'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1'Q"1€'Q"1Å‘"1üÅ‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1÷4#1,÷4#1#�#1˜#�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyМ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UA�Ê  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyМ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UA�Ê  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyМ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UA�Ê   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 œ¼AaHh bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1•Õ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1TÔ)©þ~ÓSõ 5 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>/?/I¤I¤ST^ˆ^ˆhùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbçbçdZeZgÍgÍi@j@l³l³n$o$q•q•stvwvwxèxèzY{Y}Ê}Ê;€;‚¬‚¬„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1é}!1Ðé}!1‡¾!1L‡¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1µç!1(µç!1S("1¤S("1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1‡Q"1€‡Q"1%’"1ü%’"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#15#1,5#1ï�#1˜ï�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™5 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 F†Ç”Ô bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™5 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 F†Ç”Ô bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™5  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 F†Ç”Ô  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ™¹]}Bb bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1±ñ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1”uõVÖ7·Á €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜQQÆÆ);*;4°4°>%?%I˜I˜S T ^|^|hîhîrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒvvëë` `"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1‰}!1Љ}!1'¾!1L'¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1[ç!1([ç!1ù'"1¤ù'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1-Q"1€-Q"1Ë‘"1üË‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1i2#1,i2#1A�#1˜A�#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1A�&1ÌA�&1S'1S'1¹e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÓŸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @B‚Í  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÓŸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @B‚Í  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÓŸ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @B‚Í   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‰©IiBb bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€à`À@  P €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝUUÍÍ)E*E4½4½>5?5I«I«S!T!^—^—h i sƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%���� } }"õ"õ$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=>@•@•B C E…E…GýGýIuJuLëLëNaOaQ×Q×SMTMVÃVÃX9Y9[¯[¯]%^%`›`›bce‡e‡gýgýisjslélén_o_qÕqÕsKtKvÁvÁx7y7{­{­}#~#€™€™‚ƒ…………‡û‡û‰qŠqŒçŒçŽ]�]‘Ó‘Ó“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1Ý}!1ÐÝ}!1{¾!1L{¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1¯ç!1(¯ç!1M("1¤M("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1�Q"1€�Q"1’"1ü’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1i2#1,i2#1)�#1˜)�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1ù€&1Ìù€&1ÉR'1ÉR'1qe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ&ò b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 JŠË £ã bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ&ò b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 JŠË £ã bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ&ò  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 JŠË £ã  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 †¦<\a� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1C€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ƒâbÁA  ó 3 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>4?4I«I«S"T"^”^”hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãI\J\LÓLÓNJOJQÁQÁS8T8V¯V¯X&Y&[�[�]^`‹`‹bceyeygðgðigjglÞlÞnPoPqÂqÂs4t4v¦v¦xy{Š{Š}ü}ün€n‚à‚à„R…R‡ćĉ6Š6Œ¨Œ¨Ž�‘Œ‘Œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1‰}!1Љ}!1'¾!1L'¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Uç!1(Uç!1ó'"1¤ó'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1'Q"1€'Q"1Å‘"1üÅ‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1 5#1, 5#1ã�#1˜ã�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÑ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @B‚Î bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÑ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @B‚Î bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÑ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @B‚Î  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �°Dd/O bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 à€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€`à@À  °ð €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÇÇ)=*=4³4³>)?)I�I�ST^…^…hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒuŒuŽéŽé�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1ã}!1Ðã}!1�¾!1L�¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1¯ç!1(¯ç!1M("1¤M("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1�Q"1€�Q"1’"1ü’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1 5#1, 5#1Å�#1˜Å�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Ÿ€&1ÌŸ€&1oR'1oR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ*ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I‰Ê £ã bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ*ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I‰Ê £ã bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ*ö  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I‰Ê £ã  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ž®,L`€ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¢â€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÿ^Þ=½œ’Ò €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Û ÛPPÅÅ):*:4¯4¯>&?&I›I›ST^‚^‚hôhôrfsf} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÛÛPPÅÅ : : ¯ ¯$$™™ƒƒøømmââW W"Ì"Ì$A%A'¶'¶)+*+, , ./1Š1Š3ÿ3ÿ5t6t8é8é:^;^=Ó=Ó?H@HB½B½D2E2G§G§IJL“L“NOQ}Q}SòSòUgVgXÜXÜZQ[Q]Æ]Æ_;`;b°b°d%e%gšgšijl„l„nönöphqhsÚsÚuLvLx¾x¾z0{0}¢}¢€‚†‚†„ø„ø†j‡j‰܉܋NŒNŽÀŽÀ�2‘2“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1�}!1Ð�}!1-¾!1L-¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1aç!1(aç!1ÿ'"1¤ÿ'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"13Q"1€3Q"1Ñ‘"1üÑ‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1o2#1,o2#1û�#1˜û�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ô  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  KA�Ï bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ô  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  KA�Ï bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ô   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  KA�Ï  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 € Bb2R bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1A€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�ìl×WÂBÁ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>/?/I¤I¤ST^Ž^Žhisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbébéd^e^gÓgÓiHjHl½l½n2o2q§q§stv‘v‘xy{{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1×}!1Ð×}!1u¾!1Lu¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1£ç!1(£ç!1A("1¤A("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1uQ"1€uQ"1’"1ü’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1i2#1,i2#1•�#1˜•�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ç b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M�ÎžÞ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ç b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M�ÎžÞ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ç  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M�ÎžÞ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ÿ.NZz bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€çgÎNµ5`  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll æ æ``ÖÖ)L*L4Â4Â>8?8I¬I¬S T ^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllææ``ÚÚ T T Î ÎHHÂÂ<<¶¶00ªª$$ ž ž"#%Ž%Ž'(*z*z,ð,ð.f/f1Ü1Ü3R4R6È6È8>9>;´;´=*>*@ @ BCEŒEŒGHJxJxLìLìN`O`QÔQÔSHTHV¼V¼X0Y0[¤[¤]^`Œ`Œbcetetgègèi\j\lÐlÐnDoDq¸q¸s,t,v v xy{ˆ{ˆ}ü}üp€p‚ä‚ä„X…X‡̇̉@Š@Œ´Œ´Ž(�(‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1�}!1Ð�}!1-¾!1L-¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1[ç!1([ç!1ù'"1¤ù'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1-Q"1€-Q"1Ë‘"1üË‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#15#1,5#1£Ž#1˜£Ž#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Ÿ€&1ÌŸ€&1oR'1oR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Òž b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JA�Ð bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Òž b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JA�Ð bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Òž  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JA�Ð  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‚¢Ee4 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¬ì€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1­-˜ƒnîl¬ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜRRÇÇ)<*<4±4±>&?&I™I™S T ^^hòhòrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒxxîîd d"Ú"Ú$O%O'Ä'Ä)9*9,®,®.#/#1˜1˜3 4 6‚6‚8÷8÷:l;l=á=á?V@VBËBËD@E@GµGµI*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b¨b¨degŽgŽijltltnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1×}!1Ð×}!1u¾!1Lu¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1©ç!1(©ç!1G("1¤G("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1{Q"1€{Q"1’"1ü’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#15#1,5#1=Ž#1˜=Ž#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒë b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 LŒÍ Ÿß bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒë b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 LŒÍ Ÿß bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒë  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 LŒÍ Ÿß  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ƒ£6a� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1§ç€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1œƒjêQÑG €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âZZÎÎ)B*B4¶4¶>*?*IœIœST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢’’  ‚ ‚"ú"ú$n%n'â'â)V*V,Ê,Ê.>/>1²1²3&4&6š6š89;‚;‚=ö=ö?j@jBÞBÞDRERGÆGÆI:J:L¬L¬NOQ�Q�STVtVtXæXæZX[X]Ê]Ê_<`v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ë}!1�Ë}!1i¾!1i¾!1³Â!1$³Â!1�ç!1��ç!1;("1;("1…,"1$…,"1oQ"1�oQ"1 ’"1 ’"1W–"1$W–"1óØ"1�óØ"1Sý"1üSý"1/0#1 /0#1û~#1| û~#1á$1Ü á$1¹ò$1  ¹ò$1óÞ%1¤ óÞ%1ÿ€&1„ ÿ€&1ÏR'1ÈÏR'1we'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1-aY�æ²Û b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1-aY�æ²Û b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1-aY�æ²Û b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10zz ô ôjjàà)V*V4Ê4Ê>>?>I²I²S&T&^š^šhis€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðzzôônnèè b b Ü ÜVVÐÐJJÄÄ::°°&& œ œ"#%ˆ%ˆ'þ'þ)t*t,ê,ê.`/`1Ö1Ö3L4L6Â6Â8898;®;®=">"@–@–B C E~E~GòGòIfJfLÚLÚNNONQÂQÂS6T6VªVªXY[’[’]^`z`zbîbîdbebgÖgÖiJjJl¾l¾n2o2q¦q¦stvŽvŽxy{v{v}ê}ê^€^‚Ò‚Ò„F…F‡¸‡¸‰+Š+ŒžŒžŽ�‘„‘„“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ñ}!1�Ñ}!1o¾!1 o¾!1_Å!1_Å!1£ç!1€£ç!1A("1üA("11/"11/"1uQ"1puQ"1’"1ì’"1™"1ô™"1óØ"1`óØ"1Sý"1ÌSý"1/0#1ì/0#1û~#1L û~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÛ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¯ï€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÛ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¯ï€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÛ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¯ï €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr è è]]ÒÒ)G*G4º4º>-?-I I ST^†^†hùhùrjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrèè^^ÔÔ J J À À66¬¬""˜˜  ‚‚÷÷l l"á"á$V%V'Ë'Ë)@*@,µ,µ.*/*1Ÿ1Ÿ346‰6‰8þ8þ:s;s=æ=æ?Y@YBÌBÌD?E?G²G²I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b£b£deg‰g‰iüiükolonânâpUqUsÈsÈu;v;x®x®z!{!}”}”€‚z‚z„í„í†^‡^‰ЉЋBŒBŽ´Ž´�&‘&“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ñ}!1�Ñ}!1o¾!1o¾!1³Â!1$³Â!1�ç!1��ç!1;("1;("1…,"1$…,"1oQ"1�oQ"1 ’"1 ’"1W–"1$W–"1óØ"1�óØ"1Sý"1üSý"1/0#1 /0#1û~#1| û~#1É$1Ü É$1¡ò$1  ¡ò$1ÛÞ%1¤ ÛÞ%1‡€&1„ ‡€&1WR'1ÈWR'1ÿd'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1.bW‹â®Þ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1.bW‹â®Þ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1.bW‹â®Þ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx ð ðddØØ)L*L4¾4¾>0?0I¢I¢ST^†^†høhørhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðxxððhhàà X X Ð ÐHHÀÀ88°°$$˜˜  € €"ô"ô$h%h'Ü'Ü)P*P,Ä,Ä.8/81¬1¬3 4 6”6”89;|;|=î=î?`@`BÒBÒDDEDG¶G¶I(J(LšLšN O Q~Q~SðSðUbVbXÔXÔZF[F]¸]¸_*`*bœbœdeg€g€iòiòkdldnÖnÖpHqHsºsºu,v,xžxžz{}‚}‚ôô�f‚f„؄؆H‡H‰¹‰¹‹*Œ*ޛޛ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ñ}!1�Ñ}!1o¾!1 o¾!1_Å!1_Å!1£ç!1€£ç!1A("1üA("11/"11/"1uQ"1puQ"1’"1ì’"1™"1ô™"1óØ"1`óØ"1Sý"1ÌSý"1/0#1ì/0#1û~#1L û~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1?€&1T ?€&1R'1˜R'1·d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÝ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¯ï€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÝ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¯ï€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1AuÝ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¯ï €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ê ê]]ÐÐ)C*C4´4´>%?%I–I–ST^x^xhéhérXsX} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttêê``ÖÖ L L  Â88®®$$šš  €€óóf f"Ù"Ù$L%L'¿'¿)2*2,¥,¥./1‹1‹3þ3þ5q6q8ä8ä:W;W=È=È?9@9BªBªDEGŒGŒIýIýKnLnNßNßPPQPSÁSÁU2V2X£X£Z[]…]…_ö_öagbgdØdØfIgIiºiºk+l+nœnœp q s~s~uïuïw`x`zÑzÑ|B}B³³�$‚$„•„•†‡‰t‰t‹ä‹ä�TŽT�Ä�Ä’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1¡}!1�¡}!1?¾!1 ?¾!1¹Â!1¹Â!1sç!1€sç!1("1ü("1‹,"1‹,"1EQ"1pEQ"1ã‘"1ìã‘"1]–"1ô]–"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1G�&1T G�&1S'1˜S'1¿e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠã¯ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠã¯ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠã¯ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ï ïhháá)Z*Z4Ó4Ó>J?JIÁIÁS6T6^©^©his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvïïhháá Z Z Ó ÓLLÅÅ>>··00©©"" › ›"#%�%�'(**,ø,ø.q/q1ê1ê3c4c6Ü6Ü8U9U;Î;Î=G>G@À@ÀB9C9E²E²G+H+J¢J¢LMO�O�QRT~T~VõVõXlYl[ã[ã]Z^Z`Ñ`ÑbFcFe»e»g0h0j¥j¥lmo�o�qrtstsvævæxYyY{Ì{Ì}?~?€²€²‚%ƒ%…˜…˜‡ ˆ Š{Š{ŒíŒíŽ_�_‘ёѓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1×}!1�×}!1u¾!1 u¾!1MÅ!1MÅ!1©ç!1€©ç!1G("1üG("1/"1/"1{Q"1p{Q"1’"1ì’"1ñ˜"1ôñ˜"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ä b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1žÞ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ä b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1žÞ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ä b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1žÞ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq é é^^ÓÓ)H*H4½4½>1?1I¥I¥ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqééaaÙÙ Q Q É ÉAA¹¹11©©““ } }"ò"ò$g%g'Ü'Ü)Q*Q,Æ,Æ.;/;1°1°3%4%6š6š89;„;„=ù=ù?n@nBãBãDXEXGÍGÍIAJALµLµN)O)Q�Q�STV…V…XùXùZm[m]á]á_U`UbÉbÉd=e=g±g±i%j%l™l™n o q�q�sõsõuivixÝxÝzQ{Q}Å}Å9€9‚­‚­„!…!‡•‡•‰ Š ŒŒŽôŽô�i‘i“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1›}!1�›}!19¾!1 9¾!1¹Â!1¹Â!1mç!1€mç!1 ("1ü ("1‹,"1‹,"1?Q"1p?Q"1Ý‘"1ìÝ‘"1]–"1ô]–"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1Õ€&1T Õ€&1¥R'1˜¥R'1Me'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆß« b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆß« b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆß« b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ë ëbbÙÙ)P*P4Ç4Ç>>µµ,,££‘‘  "ö"ö$m%m'ä'ä)[*[,Ò,Ò.I/I1À1À37476®6®8%9%;œ;œ=>@Š@ŠBCExExGïGïIdJdLÙLÙNNONQÃQÃS8T8V­V­X"Y"[—[—] ^ `�`�bôbôdgeggÚgÚiMjMlÀlÀn1o1q¢q¢stv„v„xõxõzf{f}×}×H€H‚¹‚¹„*…*‡š‡š‰ Š Œ|Œ|ŽíŽí�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1×}!1�×}!1u¾!1 u¾!1MÅ!1MÅ!1©ç!1€©ç!1G("1üG("1/"1/"1{Q"1p{Q"1’"1ì’"1ñ˜"1ôñ˜"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~æ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ÿß€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~æ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ÿß€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~æ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ÿß €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ç ç\\ÑÑ)F*F4»4»>.?.I¡I¡ST^…^…h÷h÷risi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqçç]]ÓÓ I I ¿ ¿55««!!——  ��öök k"à"à$U%U'Ê'Ê)?*?,´,´.)/)1ž1ž346ˆ6ˆ8ý8ý:r;r=ç=ç?\@\BÑBÑDFEFG»G»I.J.L¡L¡NOQ‡Q‡SúSúUmVmXàXàZS[S]Æ]Æ_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰̉̋?Œ?޲޲�%‘%“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1›}!1�›}!19¾!1 9¾!1³Â!1³Â!1mç!1€mç!1 ("1ü ("1…,"1…,"1?Q"1p?Q"1Ý‘"1ìÝ‘"1W–"1ôW–"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1M�&1T M�&1S'1˜S'1Åe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡à¬ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡à¬ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡à¬ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ê êbbÚÚ)R*R4Ê4Ê>@?@I¶I¶S,T,^¢^¢hisŽsŽ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrêêbbÚÚ R R Ê ÊBBºº22ªª""šš Š Š"#%z%z'ò'ò)j*j,â,â.Z/Z1Ò1Ò3J4J6Â6Â8:9:;²;²=*>*@¢@¢BCE’E’G H J€J€LöLöNlOlQâQâSXTXVÎVÎXDYD[º[º]0^0`¦`¦bce’e’ghj~j~lôlônjojqàqàsVtVvÌvÌxByB{¸{¸}.~.€¤€¤‚ƒ…�…�‡ˆŠ|Š|ŒòŒòŽh�h‘Þ‘Þ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Ý}!1�Ý}!1{¾!1 {¾!1_Å!1_Å!1©ç!1€©ç!1G("1üG("11/"11/"1{Q"1p{Q"1’"1ì’"1™"1ô™"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡!í b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1·÷€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡!í b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1·÷€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡!í b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1·÷ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ä ä[[ÒÒ)I*I4À4À>7?7I®I®S%T%^—^—h i s{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmää[[ÒÒ I I À À77®®%%œœŠŠ x x"ï"ï$f%f'Ý'Ý)T*T,Ë,Ë.B/B1¹1¹30406§6§89;•;•= > @ƒ@ƒBúBúDqEqGèGèI_J_LÖLÖNMOMQÄQÄS;T;V²V²X)Y)[ [ ]^`Ž`Žbce|e|gógóijjjlálánSoSqÅqÅs7t7v©v©xy{�{�}ÿ}ÿq€q‚ã‚ã„U…U‡LJlj9Š9Œ«Œ«Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1•}!1�•}!13¾!1 3¾!1³Â!1³Â!1gç!1€gç!1("1ü("1…,"1…,"19Q"1p9Q"1ב"1ìב"1W–"1ôW–"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Õ€&1T Õ€&1¥R'1˜¥R'1Me'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ܨ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ܨ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ܨ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æ\\ÒÒ)H*H4¾4¾>2?2I¦I¦ST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  ‚‚øøn n"ä"ä$Z%Z'Ð'Ð)F*F,¼,¼.2/21¨1¨346”6”8 9 ;€;€=ö=ö?l@lBâBâDXEXGÎGÎIBJBL¶L¶N*O*QžQžSTV†V†XúXúZn[n]â]â_V`VbÊbÊd>e>g²g²i&j&lšlšnoq‚q‚sösöujvjxÞxÞzR{R}Æ}Æ:€:‚®‚®„"…"‡–‡–‰ Š Œ~Œ~ŽòŽò�f‘f“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Ý}!1�Ý}!1{¾!1 {¾!1_Å!1_Å!1¯ç!1€¯ç!1M("1üM("11/"11/"1�Q"1p�Q"1’"1ì’"1™"1ô™"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡#ï b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¶ö€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡#ï b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¶ö€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡#ï b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¶ö €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm â âWWÌÌ)A*A4¶4¶>+?+I I ST^‡^‡hùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmââWWÌÌ A A ¶ ¶++  ŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?O@OBÄBÄD9E9G®G®I#J#L˜L˜N O Q‚Q‚S÷S÷UlVlXáXáZV[V]Ë]Ë_@`@bµbµd*e*gŸgŸijl‰l‰nûnûpmqmsßsßuQvQxÃxÃz5{5}§}§€‚‹‚‹„ý„ý†o‡o‰á‰á‹SŒSŽÅŽÅ�7‘7“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1•}!1�•}!13¾!1 3¾!1ÅÂ!1ÅÂ!1aç!1€aç!1ÿ'"1üÿ'"1—,"1—,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1i–"1ôi–"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1 �&1T  �&1ÛR'1˜ÛR'1ƒe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„Ý© b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1"b€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„Ý© b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1"b€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„Ý© b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1"b €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn å å\\ÓÓ)J*J4Á4Á>6?6I«I«S T ^•^•h i ss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnåå\\ÓÓ J J Á Á88¯¯&&��‹‹ y y"ð"ð$g%g'Þ'Þ)U*U,Ì,Ì.C/C1º1º31416¨6¨89;–;–= > @„@„BûBûDrErGéGéI^J^LÓLÓNHOHQ½Q½S2T2V§V§XY[‘[‘]^`{`{bðbðdeeegÚgÚiOjOlÄlÄn9o9q®q®s#t#v˜v˜x y {‚{‚}÷}÷l€l‚á‚á„V…V‡ˇˉ@Š@ŒµŒµŽ*�*‘Ÿ‘Ÿ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1ã}!1�ã}!1�¾!1 �¾!15Å!15Å!1¯ç!1€¯ç!1M("1üM("1/"1/"1�Q"1p�Q"1’"1ì’"1Ù˜"1ôÙ˜"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�*ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ï€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�*ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ï€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�*ö b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ï €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ç çaa××)M*M4Ã4Ã>7?7I«I«ST^“^“his{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmççaaÛÛ U U Ï ÏIIÃÃ==··11««%% Ÿ Ÿ"#%�%�'(*{*{,ñ,ñ.g/g1Ý1Ý3S4S6É6É8?9?;µ;µ=+>+@¡@¡BCE�E�GHJwJwLëLëN_O_QÓQÓSGTGV»V»X/Y/[£[£]^`‹`‹bÿbÿdsesgçgçi[j[lÏlÏnCoCq·q·s+t+vŸvŸxy{‡{‡}û}ûo€o‚ã‚ã„W…W‡ˇˉ?Š?Œ³Œ³Ž'�'‘›‘›“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1�}!1��}!1-¾!1 -¾!1ËÂ!1ËÂ!1aç!1€aç!1ÿ'"1üÿ'"1�,"1�,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1o–"1ôo–"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚Ù¥ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1#c€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚Ù¥ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1#c€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚Ù¥ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1#c €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ã ãYYÎÎ)C*C4¸4¸>+?+IžIžST^„^„h÷h÷rjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““  õõk k"á"á$V%V'Ë'Ë)@*@,µ,µ.*/*1Ÿ1Ÿ346‰6‰8þ8þ:s;s=è=è?]@]BÒBÒDGEGG¼G¼I/J/L¢L¢NOQˆQˆSûSûUnVnXáXáZT[T]Ç]Ç_:`:b­b­d e g“g“ijlylynìnìp_q_sÒsÒuEvEx¸x¸z+{+}ž}ž€‚„‚„„÷„÷†j‡j‰݉݋PŒPŽÃŽÃ�6‘6“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1ã}!1�ã}!1�¾!1 �¾!15Å!15Å!1µç!1€µç!1S("1üS("1/"1/"1‡Q"1p‡Q"1%’"1ì%’"1Ù˜"1ôÙ˜"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�,ø b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ï€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�,ø b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ï€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�,ø b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ï €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±Ñ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm å å]]ÑÑ)E*E4¹4¹>+?+I�I�ST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmåå]]ÕÕ M M Å Å==µµ--¥¥••  … …"ý"ý$q%q'å'å)Y*Y,Í,Í.A/A1µ1µ3)4)6�6�89;…;…=ù=ù?m@mBáBáDUEUGÉGÉI;J;L­L­NOQ‘Q‘STVuVuXçXçZY[Y]Ë]Ë_=`=b¯b¯d!e!g“g“ijlwlwnénép[q[sÍsÍu?v?x±x±z#{#}•}•€‚y‚y„ë„ë†]‡]‰ωϋAŒA޳޳�%‘%“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1�}!1��}!1-¾!1 -¾!1³Â!1³Â!1[ç!1€[ç!1ù'"1üù'"1…,"1…,"1-Q"1p-Q"1Ë‘"1ìË‘"1W–"1ôW–"1aÚ"1`aÚ"1Áþ"1ÌÁþ"10#1ì0#1é~#1L é~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1á€&1T á€&1±R'1˜±R'1Ye'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Ú¦ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Ú¦ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Ú¦ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1.N  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn è è^^ÔÔ)J*J4¾4¾>2?2I¦I¦ST^Ž^Žhistst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnèèbbÜÜ V V Ð ÐJJÄÄ>>¸¸..¤¤ � �"#%|%|'ò'ò)h*h,Þ,Þ.T/T1Ê1Ê3@4@6¶6¶8,9,;¢;¢=>@Š@ŠBþBþDrErGæGæIZJZLÎLÎNBOBQ¶Q¶S*T*VžVžXY[†[†]ú]ú_n`nbâbâdVeVgÊgÊi>j>l²l²n&o&qšqšstv‚v‚xöxözj{j}Þ}ÞR€R‚ƂƄ:…:‡¬‡¬‰ŠŒ’Œ’Ž�‘x‘x“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ã}!1�ã}!1�¾!1 �¾!1SÅ!1SÅ!1µç!1€µç!1S("1üS("1%/"1%/"1‡Q"1p‡Q"1%’"1ì%’"1÷˜"1ô÷˜"1gÚ"1`gÚ"1Çþ"1ÌÇþ"10#1ì0#1é~#1L é~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™3ÿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1´ô€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™3ÿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1´ô€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™3ÿ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1´ô €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations14T  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜQQÆÆ);*;4®4®>!?!I”I”ST^z^zhíhír^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒvvëë` `"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ú=Ú?M@MBÀBÀD3E3G¦G¦IJLŒLŒNÿNÿPrQrSåSåUXVXXËXËZ>[>]±]±_$`$b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†R‡R‰ĉċ6Œ6ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1‰}!1�‰}!1'¾!1 '¾!1³Â!1³Â!1[ç!1€[ç!1ù'"1üù'"1…,"1…,"1-Q"1p-Q"1Ë‘"1ìË‘"1W–"1ôW–"1sÚ"1`sÚ"1Óþ"1ÌÓþ"10#1ì0#1é~#1L é~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1i€&1T i€&19R'1˜9R'1ád'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÖ¢ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÖ¢ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÖ¢ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Aa  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ä äXXÌÌ)@*@4²4²>$?$I–I–ST^z^zhìhìr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllää\\ÔÔ L L Ä Ä<<´´,,¤¤ŒŒ t t"è"è$\%\'Ð'Ð)D*D,¸,¸.,/,1 1 346ˆ6ˆ8ü8ü:p;p=â=â?T@TBÆBÆD8E8GªGªIJLŽLŽNOQrQrSäSäUVVVXÈXÈZ:[:]¬]¬_`b�b�degtgtiæiækXlXnÊnÊp?IŠIŠSûSû]l^lhÝhÝrLsL} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽttççZ Z"Í"Í$@%@'³'³)&*&,™,™. / 113ò3ò5e6e8Ø8Ø:K;K=¼=¼?-@-BžBžDEG€G€IñIñKbLbNÓNÓPDQDSµSµU&V&X—X—Z[]y]y_ê_êa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1ƒ}!1�ƒ}!1!¾!1 !¾!1¿Â!1¿Â!1Uç!1€Uç!1ó'"1üó'"1‘,"1‘,"1'Q"1p'Q"1Å‘"1ìÅ‘"1c–"1ôc–"1aÚ"1`aÚ"1Áþ"1ÌÁþ"10#1ì0#1ã~#1L ã~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1)�&1T )�&1ùR'1˜ùR'1¡e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~×£ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1_€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~×£ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1_€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~×£ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1_ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations15U  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ã ã\\ÕÕ)N*N4Ç4Ç>>?>IµIµS*T*^�^�his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjãã\\ÕÕ N N Ç Ç@@¹¹22««$$�� � �"#%�%�'ú'ú)s*s,ì,ì.e/e1Þ1Þ3W4W6Ð6Ð8I9I;Â;Â=;>;@´@´B-C-E¦E¦GHJ–J–L M O„O„QûQûSrTrVéVéX`Y`[×[×]N^N`Å`Åb:c:e¯e¯g$h$j™j™lmo�o�qôqôsgtgvÚvÚxMyM{À{À}3~3€¦€¦‚ƒ…Œ…Œ‡ý‡ý‰oŠoŒáŒáŽS�S‘őœ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1é}!1�é}!1‡¾!1 ‡¾!1/Å!1/Å!1»ç!1€»ç!1Y("1üY("1/"1/"1�Q"1p�Q"1+’"1ì+’"1Ó˜"1ôÓ˜"1gÚ"1`gÚ"1Çþ"1ÌÇþ"10#1ì0#1ã~#1L ã~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1À&1T À&1“R'1˜“R'1;e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢< b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ï€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢< b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ï€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢< b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ï €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1;[  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝRRÇÇ)<*<4±4±>%?%I™I™S T ^�^�hõhõrisi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��‡‡üüq q"æ"æ$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`Ib½b½d1e1g¥g¥ijl�l�noququséséu]v]xÑxÑzE{E}¹}¹-€-‚¡‚¡„…‡‰‡‰‰þ‰þ‹sŒsŽèŽè�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1ƒ}!1�ƒ}!1!¾!1 !¾!1¿Â!1¿Â!1Oç!1€Oç!1í'"1üí'"1‘,"1‘,"1!Q"1p!Q"1¿‘"1ì¿‘"1c–"1ôc–"1yÚ"1`yÚ"1Ùþ"1ÌÙþ"10#1ì0#1ã~#1L ã~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1·€&1T ·€&1‡R'1˜‡R'1/e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|ÓŸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 `€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|ÓŸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 `€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|ÓŸ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 ` €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ll  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>0?0I¥I¥ST^‰^‰húhúrjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n%o%q–q–stvxvxxéxézZ{Z}Ë}Ë<€<‚­‚­„…‡އމÿ‰ÿ‹pŒpŽáŽá�R‘R“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1ï}!1�ï}!1�¾!1 �¾!1/Å!1/Å!1»ç!1€»ç!1Y("1üY("1/"1/"1�Q"1p�Q"1+’"1ì+’"1Ó˜"1ôÓ˜"1…Ú"1`…Ú"1åþ"1Ìåþ"10#1ì0#1ã~#1L ã~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1{€&1T {€&1KR'1˜KR'1ód'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢>  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ï€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢>  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ï€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢>  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ï €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Vv  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛPPÅÅ):*:4¯4¯>"?"I•I•ST^y^yhëhër]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹uuêê_ _"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I"J"L•L•NOQ{Q{SîSîUaVaXÔXÔZG[G]º]º_-`-bŸbŸdegƒgƒiõiõkglgnÙnÙpKqKs½s½u/v/x¡x¡z{}…}…÷÷�i‚i„Û„Û†M‡M‰À‰À‹3Œ3ަަ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1}}!1�}}!1¾!1 ¾!1§Â!1§Â!1Oç!1€Oç!1í'"1üí'"1y,"1y,"1!Q"1p!Q"1¿‘"1ì¿‘"1K–"1ôK–"1aÚ"1`aÚ"1Áþ"1ÌÁþ"10#1ì0#1ã~#1L ã~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%15�&1T 5�&1S'1˜S'1­e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ô  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ô  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ô  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations18X  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÎÎ)F*F4¾4¾>4?4IªIªS T ^˜^˜his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$n%n'æ'æ)^*^,Ö,Ö.N/N1Æ1Æ3>4>6¶6¶8.9.;¦;¦=>@–@–BCE†E†GþGþItJtLêLêN`O`QÖQÖSLTLVÂVÂX8Y8[®[®]$^$`š`šbce†e†gügüirjrlèlèn`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1ï}!1�ï}!1�¾!1 �¾!1MÅ!1MÅ!1Áç!1€Áç!1_("1ü_("1/"1/"1“Q"1p“Q"11’"1ì1’"1ñ˜"1ôñ˜"1CÚ"1`CÚ"1£þ"1Ì£þ"10#1ì0#1ã~#1L ã~#1Ã$1¬ Ã$1ëð$1p ëð$1ÉÀ%1t ÉÀ%1í€&1T í€&1ãP'1˜ãP'1‹c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«E b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1±ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«E b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1±ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«E b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1±ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1@<  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>+?+I¢I¢ST^•^•hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜISJSLÊLÊNAOAQ¸Q¸S/T/V¦V¦XY[”[”] ^ `‚`‚bùbùdpepgçgçi^j^lÕlÕnQoQqÃqÃs5t5v§v§xy{‹{‹}ý}ýo€o‚á‚á„S…S‡Ňʼn7Š7Œ©Œ©Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1}}!1�}}!1¾!1 ¾!1§Â!1§Â!1Iç!1€Iç!1ç'"1üç'"1y,"1y,"1Q"1pQ"1¹‘"1칑"1K–"1ôK–"1Ú"1`Ú"1ßþ"1Ìßþ"10#1ì0#1ã~#1L ã~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1½€&1T ½€&1�R'1˜�R'15e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyМ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyМ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyМ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Yy  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÆÆ)<*<4²4²>&?&IšIšST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$N%N'Ä'Ä):*:,°,°.&/&1œ1œ346ˆ6ˆ8þ8þ:t;t=ê=ê?`@`BÖBÖDLELGÂGÂI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1ï}!1�ï}!1�¾!1 �¾!1GÅ!1GÅ!1Áç!1€Áç!1_("1ü_("1/"1/"1“Q"1p“Q"11’"1ì1’"1ë˜"1ôë˜"1gÚ"1`gÚ"1Çþ"1ÌÇþ"10#1ì0#1ã~#1L ã~#1±$1¬ ±$1Ùð$1p Ùð$1·À%1t ·À%1“€&1T “€&1‰P'1˜‰P'11c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«G b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1°ð€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«G b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1°ð€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«G b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1°ð €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1c_  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ö ÖKKÀÀ)5*54ª4ª>?I”I”S T ^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÖÖKKÀÀ 5 5 ª ª””  ~~óóhhÝÝR R"Ç"Ç$<%<'±'±)&*&,›,›./1…1…3ú3ú5o6o8ä8ä:Y;Y=Î=Î?C@CB¸B¸D-E-G¢G¢IJLŒLŒNOQvQvSëSëU`V`XÕXÕZJ[J]¿]¿_4`4b©b©deg“g“ijl}l}n÷n÷piqisÛsÛuMvMx¿x¿z1{1}£}£€‚‡‚‡„ù„ù†k‡k‰݉݋OŒOŽÁŽÁ�3‘3“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1w}!1�w}!1¾!1 ¾!1³Â!1³Â!1Iç!1€Iç!1ç'"1üç'"1…,"1…,"1Q"1pQ"1¹‘"1칑"1W–"1ôW–"1±Ø"1`±Ø"1ý"1Ìý"10#1€0#1Ý~#1àÝ~#1É$1@É$1ñð$1Ú ñð$1ÏÀ%1ú ÏÀ%1ÿ€&1Ò ÿ€&1õP'1r õP'1�c'1’ €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÑ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1\€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÑ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1\€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÑ� b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1\ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±­  bÒ-g�operation.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 :z2â ú reasonOperatorInputsvalue8¢Planned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })binvocations1Ç0bb Ù ÙPPÉÉ)@*@4·4·>,?,I¡I¡ST^—^—h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbæbæd[e[gÐgÐiEjElºlºn;o;q°q°s%t%všvšxy{„{„}ù}ùn€n‚ã‚ã„X…X‡͇͉BŠBŒ·Œ·Ž,�,‘¡‘¡“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1õ}!1�õ}!1“¾!1 “¾!1#Å!1#Å!1Çç!1€Çç!1e("1üe("1õ."1õ."1™Q"1p™Q"17’"1ì7’"1ǘ"1ôǘ"1±Ø"1`±Ø"1ý"1Ìý"10#1ì0#1Ý~#1L Ý~#1É$1¬ É$1ñð$1p ñð$1ÏÀ%1t ÏÀ%1ÿ€&1T ÿ€&1õP'1˜õP'1�c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´N b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ï€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´N b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ï€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´N b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ï €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±­  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Û ÛUUÕÕ)K*K4Á4Á>5?5I©I©ST^�^�his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÛÛUUÏÏ I I à Ã==··11««%%ŸŸ “ “" # %�%�'(*y*y,ï,ï.e/e1Û1Û3Q4Q6Ç6Ç8=9=;³;³=)>)@Ÿ@ŸBCE‹E‹GHJuJuLéLéN]O]QÑQÑSETEV¹V¹X-Y-[¡[¡]^`‰`‰býbýdqeqgågåiYjYlÍlÍnMoMqÁqÁs5t5v©v©xy{‘{‘}~€y€y‚í‚í„a…a‡Õ‡Õ‰IŠIŒ½Œ½Ž1�1‘¥‘¥“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1q}!1�q}!1¾!1 ¾!1³Â!1³Â!1Cç!1€Cç!1á'"1üá'"1…,"1…,"1Q"1pQ"1³‘"1쳑"1W–"1ôW–"1±Ø"1`±Ø"1ý"1Ìý"10#1ì0#1Ý~#1L Ý~#1±$1¬ ±$1Ùð$1p Ùð$1·À%1t ·À%1™€&1T ™€&1�P'1˜�P'17c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÍ™ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1]€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÍ™ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1]€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1BvÍ™ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1] €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±­  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×MMÆÆ);*;4°4°>#?#I–I–S T ^ˆ^ˆhûhûrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnðnðpcqcsÖsÖuIvIx¼x¼z/{/}¢}¢€‚ˆ‚ˆ„û„û†n‡n‰á‰á‹TŒTŽÇŽÇ�:‘:“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1õ}!1�õ}!1“¾!1 “¾!1#Å!1#Å!1Çç!1€Çç!1e("1üe("1õ."1õ."1™Q"1p™Q"17’"1ì7’"1ǘ"1ôǘ"1±Ø"1`±Ø"1ý"1Ìý"10#1ì0#1Ý~#1L Ý~#1±$1¬ ±$1Ùð$1p Ùð$1·À%1t ·À%1‡€&1T ‡€&1}P'1˜}P'1%c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´P b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ï€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´P b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ï€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´P b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ï €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1±­  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ù ÙQQÍÍ)A*A4µ4µ>'?'I™I™S T ^‰^‰hûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A ¹ ¹11©©!!™™‰‰ y y"ñ"ñ$m%m'á'á)U*U,É,É.=/=1±1±3%4%6™6™8 9 ;�;�=õ=õ?i@iBÝBÝDQEQGÅGÅI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnñnñpcqcsÕsÕuGvGx¹x¹z+{+}�}�€‚�‚�„ó„ó†e‡e‰׉׋IŒI޻޻�-‘-“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„%i!1G}!1G}!1å½!1|å½!1IÃ!1„IÃ!1Æ!1ðÆ!1 Ð!1| Ð!1?Ó!1H?Ó!1ç!1Ôç!1±'"1P±'"1-"1X-"1Q0"1ÄQ0"1Û9"1PÛ9"1="1="1åP"1¨åP"1ƒ‘"1$ ƒ‘"1í–"1, í–"1#š"1˜ #š"1­£"1$ ­£"1ã¦"1ð ã¦"1%Ô"1| %Ô"17Ú"1X7Ú"1—þ"1Ô—þ"1Ù+#1ôÙ+#1£:#1У:#1_o#1\_o#1±�#1€±�#1¯š#1¬¯š#1«$18«$1ƒò$1üƒò$1½Þ%1½Þ%17ì%1à7ì%1�€&1l�€&1QR'1°QR'1ùd'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P«w b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1³ó bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P«w b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1³ó  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P«w  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1³ó bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �¡³Ó? bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1c£Fs³«ëP�  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dcãFÆ)© ŒïoÒRµ5˜-­}ý Œbâëk G Ç Ê J , ¬  ‘ ö v Û [ À @£ãF†e¥ë+�Ð bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1077 Ó ÓRRÇÇ)<*<4±4±>$?$I—I—ST^‰^‰hühüroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð77­­##™™   « «!!——  ƒƒwwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?P@PBÃBÃD6E6G©G©IJL�L�NOQuQuSèSèU[V[XÎXÎZA[A]´]´_'`'b¦b¦degŒgŒiÿiÿkrlrnånåpXqXsËsËu>v>x±x±z${$}—}— € ‚}‚}„ð„ð†c‡c‰Õ‰Õ‹GŒG޹޹�+‘+“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1 ~!1è ~!1«¾!1d«¾!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!1Ùç!1@Ùç!1w("1¼w("1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1«Q"1˜«Q"1I’"1I’"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"1mÚ"1Ì mÚ"1Íþ"1H Íþ"1,#1h,#1©:#1D©:#1eo#1Ðeo#1Ã�#1ôÃ�#1Áš#1 Áš#1½$1¬½$1•ò$1p•ò$1ÏÞ%1tÏÞ%1Iì%1TIì%1Õ€&1àÕ€&1¥R'1$¥R'1Me'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐl8 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Cƒ™ÙÆ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐl8 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Cƒ™ÙÆ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐl8  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Cƒ™ÙÆ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ³Ó›»Rr bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1d¤H9y®îT”  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1ddäHÈ,¬�ôtØX¼<  oï8¸†�lìÊJ R Ò 8 ¸  ž  „ ê j Ð PÉ ´ôX˜þ>¤ä bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee ß ßUUÍÍ)*4¹4¹>-?-I¡I¡S!T!^•^•h i s}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯%%›› ‡ ‡"ý"ý$u%u'ë'ë)a*a,×,×.M/M1�1�346‰6‰8ÿ8ÿ:u;u=>@…@…BùBùDmEmGáGáIUJULÉLÉN=O=Q±Q±S%T%V™V™X Y [�[�]õ]õ_i`ibébéd]e]gÑgÑiEjEl¹l¹n-o-q¡q¡stv‰v‰xýxýzq{q}å}åY€Y‚͂̈́A…A‡µ‡µ‰(Š(Œ›Œ›Ž�‘�‘�“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„%i!1A}!1A}!1ß½!1|ß½!1IÃ!1„IÃ!1Æ!1ðÆ!1 Ð!1| Ð!1?Ó!1H?Ó!1ç!1Ôç!1±'"1P±'"1-"1X-"1Q0"1ÄQ0"1Û9"1PÛ9"1="1="1åP"1¨åP"1ƒ‘"1$ ƒ‘"1í–"1, í–"1#š"1˜ #š"1­£"1$ ­£"1ã¦"1ð ã¦"1%Ô"1| %Ô"1gÚ"1XgÚ"1Çþ"1ÔÇþ"1 ,#1ô ,#1c;#1Ðc;#1p#1\p#1_“#1€_“#1]ž#1¬]ž#1™$18™$1qò$1üqò$1«Þ%1«Þ%1%ì%1à%ì%1€&1l€&1ëQ'1°ëQ'1“d'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¨t b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1³ó bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¨t b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1³ó  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¨t  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1³ó bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ž®³ÓLl bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1”Ôh¨:zR  L  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d”¨(¼<ÐPädøx Œ  2 ² D Ä © ) V Ö ˆ  h è i é z ú H È Œ žD°0ĘØjª<| bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1088 Õ ÕPPÃÃ)6*64©4©>?I‹I‹ST^y^yhêhêr[s[} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð88®®$$šš   ­ ­##™™……ssææY Y"Ì"Ì$?%?'²'²)%*%,˜,˜. / 1~1~3ñ3ñ5d6d8×8×:J;J=½=½?.@.BŸBŸDEG�G�IòIòKcLcNÔNÔPEQES¶S¶U'V'X˜X˜Z [ ]z]z_ë_ëahbhdÙdÙfJgJi»i»k,l,n�n�pqssuðuðwaxazÒzÒ|C}C´´�%‚%„–„–†‡‰w‰w‹ç‹ç�WŽW�Ç�Ç’# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1 ~!1è ~!1«¾!1d«¾!1YÅ!1lYÅ!1ãÎ!1èãÎ!1Ò!1´Ò!1ßç!1@ßç!1}("1¼}("1+/"1Ä+/"1µ8"1@µ8"1ë;"1 ë;"1±Q"1˜±Q"1O’"1O’"1ý˜"1 ý˜"1‡¢"1˜ ‡¢"1½¥"1d ½¥"1ÿÒ"1ð ÿÒ"17Ú"1Ì 7Ú"1—þ"1H —þ"1Ù+#1hÙ+#1ë:#1Dë:#1§o#1Чo#1‘#1ô‘#1œ#1 œ#1¥$1¬¥$1}ò$1p}ò$1·Þ%1t·Þ%11ì%1T1ì%1c€&1àc€&13R'1$3R'1Ûd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒp< b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 CƒšÚÆ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒp< b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 CƒšÚÆ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒp<  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 CƒšÚÆ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ³Ó‡§> bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1w·.nã# ˜Ø  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dw÷nîeå\ÜSÓJÊAÁ8¸-­" ¢ “   — \ Ü  Œ % ¥  � î n ö v ë kR’à`׎ÎCƒø8 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛOOÇÇ)*4¯4¯>!?!I“I“ST^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››ƒƒ÷÷k k"ß"ß$W%W'Ë'Ë)?*?,³,³.'/'1t1t3è3è5\6\8Ð8Ð:D;D=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b™b™d e g}g}iïiïkalanÓnÓpEqEs·s·u)v)x›x›z { }}ññ�c‚c„Õ„Õ†G‡G‰¸‰¸‹)Œ)ŽšŽš� ‘ “  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1w}!1èw}!1¾!1d¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1Cç!1@Cç!1á'"1¼á'"11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1Q"1˜Q"1³‘"1³‘"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1=Ú"1Ì =Ú"1�þ"1H �þ"1ß+#1hß+#1…:#1D…:#1Ao#1ÐAo#1�#1ô�#1š#1 š#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1½€&1ཀ&1�R'1$�R'15e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwÒž b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 (hCƒÈ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwÒž b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 (hCƒÈ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwÒž  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 (hCƒÈ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 € ³Ó!A bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Zšô4�Ðl¬,l  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dZÚ4´ŽèhÂBœvöPÐ,¬œˆùyäd ÖV À @ ³ 3 œ  x ø T Ô 0 °  J¤ä@€A�Ü bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ø ØMMÂÂ)7*74¬4¬>ü>üH–I–ST^ˆ^ˆhühürrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ØØPPÈÈ @ @ ¸ ¸00¨¨  ˜˜  ‚‚÷÷l l"á"á$V%V'Ë'Ë)@*@,µ,µ.*/*1Ÿ1Ÿ346‰6‰8þ8þ:s;s=è=è?]@]BÒBÒDGEGG¼G¼I J L€L€NôNôPhQhSÜSÜUvVvXêXêZ^[^]Ò]Ò_F`FbÄbÄd8e8g¬g¬i j l”l”noq|q|sðsðudvdxØxØzL{L}À}À4€4‚¨‚¨„…‡’‡’‰ŠŒ|Œ|ŽñŽñ�f‘f“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1~!1è~!1¥¾!1d¥¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1Óç!1@Óç!1q("1¼q("11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1¥Q"1˜¥Q"1C’"1C’"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1sÚ"1Ì sÚ"1Óþ"1H Óþ"1,#1h,#1‹:#1D‹:#1Go#1ÐGo#1�#1ô�#1š#1 š#1Ã$1¬Ã$1›ò$1p›ò$1ÕÞ%1tÕÞ%1Oì%1TOì%1�&1à�&1íR'1$íR'1•e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘Åa- b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 F†•ÕÉ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘Åa- b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 F†•ÕÉ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘Åa-  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 F†•ÕÉ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ³Ó ÀXx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1[›ö68x“Ó0p  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d[Û6¶‘ìlÇG¢"}ýXØ|ü5µGÇ’“ïo Þ^ Ì L © ) †  c ã @ À [ë +¶öS“ð0 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Þ ÞWWÐÐ)I*I4Â4Â>;?;I²I²S'T'^œ^œhéhér‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÞÞWWÐÐ I I  Â;;´´--¦¦˜˜ Š Š"#%|%|'õ'õ)n*n,ç,ç.`/`1Ù1Ù3R4R6Ë6Ë8D9D;½;½=6>6@¯@¯B(C(E¡E¡GHJ“J“L M O�O�QøQøSoToVæVæX]Y][Ô[Ô]K^K`Â`Âb7c7e¬e¬g!h!j–j–l m o€o€qóqósftfvÙvÙxLyL{™{™} ~ €€‚ò‚ò„e…e‡þ‡þ‰pŠpŒâŒâŽT�T‘ƑƓ  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1q}!1èq}!1¾!1d¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1Cç!1@Cç!1á'"1¼á'"11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1Q"1˜Q"1³‘"1³‘"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1mÚ"1Ì mÚ"1Íþ"1H Íþ"1,#1h,#1:#1D:#1;o#1Ð;o#1ýŽ#1ôýŽ#1û™#1 û™#1¥$1¬¥$1}ò$1p}ò$1·Þ%1t·Þ%11ì%1T1ì%1{€&1à{€&1KR'1$KR'1ód'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwМ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 'gCƒÇ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwМ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 'gCƒÇ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1CwМ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 'gCƒÇ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‘±³ÓRr bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1[›ö6�Ï(h I  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d[Û6¶‘ìlÇG¢"}ýXØ1± Šãc ¼ < •  �  n î o ï G Ç L Ì    ) © û ;–Ö/oÈÞ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ö ÖKKÀÀ)5*54ª4ª>ø>øH’I’S T ^~^~hðhðrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††ûûppååZ Z"Ï"Ï$D%D'¹'¹).*.,£,£./1�1�346w6w8ì8ì:a;a=Ö=Ö?K@KBÀBÀD5E5GªGªIøIøKkLkNÞNÞPQQQSÄSÄU^V^XÑXÑZD[D]·]·_*`*b¤b¤degˆgˆiúiúklllnÞnÞpPqPsÂsÂu4v4x¦x¦z{}Š}Šüü�n‚n„à„à†T‡T‰ljNj:Œ:Ž­Ž­� ‘ “  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1~!1è~!1¥¾!1d¥¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1Ùç!1@Ùç!1w("1¼w("11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1«Q"1˜«Q"1I’"1I’"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1=Ú"1Ì =Ú"1�þ"1H �þ"1ß+#1hß+#1%:#1D%:#1án#1Ðán#1•�#1ô•�#1“˜#1 “˜#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1±€&1à±€&1�R'1$�R'1)e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Çe1 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G‡”ÔÊ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Çe1 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G‡”ÔÊ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Çe1  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G‡”ÔÊ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ³Óˆ¨ @ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1CƒÆI‰Ì à  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dCÆÉIŒ OÏ’ÕU˜[Ûžáa¤$ g ç © ) * ª t ô í m ? ¿ ° 0 Š s ³ ö 6 y ¹ ü <v¶ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>-?-I¢I¢ST^Š^ŠhÔhÔrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbåbådXeXgËgËi>j>l±l±n&o&q—q—stvyvyxêxêz4{4}¥}¥€‚‡‚‡„ø„ø†‘‡‘‰ŠŒsŒsŽäŽä�U‘U“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1}}!1Ð}}!1¾!1L¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Iç!1(Iç!1ç'"1¤ç'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1Q"1€Q"1¹‘"1ü¹‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1CÚ"1´ CÚ"1£þ"10 £þ"1å+#1På+#1µ7#1,µ7#11Ž#1˜1Ž#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1á€&1Ìá€&1±R'1±R'1Ye'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÓŸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TCƒÓ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÓŸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TCƒÓ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÓŸ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TCƒÓ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ÿ³Ó#C bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1[›€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ý}ØX³3 Ž  K €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` × ×NNÅÅ)<*<4³4³>,?,I£I£ST^Œ^Œhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``××NNÅÅ < < ³ ³**¡¡��}}ôôk k"â"â$Y%Y'Ð'Ð)G*G,¾,¾.5/51¬1¬3#4#6š6š89;ˆ;ˆ=ÿ=ÿ?v@vBíBíDdEdGÛGÛITJTLËLËNBOBQ¹Q¹S0T0V§V§XY[•[•] ^ `ƒ`ƒbúbúdqeqgègèi_j_lÖlÖnHoHqºqºs,t,vžvžxy{‚{‚}ô}ôf€f‚؂؄J…J‡¼‡¼‰.Š.Œ Œ Ž�‘„‘„“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1û}!1Ðû}!1™¾!1L™¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Íç!1(Íç!1k("1¤k("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1ŸQ"1€ŸQ"1=’"1ü=’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1‹7#1,‹7#1)�#1˜)�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1/�&1Ì/�&1ÿR'1ÿR'1§e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºV" b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘رñ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºV" b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘رñ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºV"  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘رñ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ³Ó¥Å^~ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations13s€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�NΛèh C €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝUUÍÍ)E*E4½4½>5?5I«I«S!T!^—^—h i sƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%���� } }"õ"õ$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=>@•@•B C E…E…GýGýIuJuLëLëNaOaQ×Q×SMTMVÃVÃX9Y9[¯[¯]%^%`›`›bce‡e‡gýgýisjslélén_o_qÕqÕsKtKvÁvÁx7y7{­{­}#~#€™€™‚ƒ…………‡û‡û‰qŠqŒçŒçŽ]�]‘Ó‘Ó“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1w}!1Ðw}!1¾!1L¾!1kÅ!1TkÅ!1õÎ!1ÐõÎ!1+Ò!1œ+Ò!1Iç!1(Iç!1ç'"1¤ç'"1=/"1¬=/"1Ç8"1(Ç8"1ý;"1ôý;"1Q"1€Q"1¹‘"1ü¹‘"1™"1 ™"1™¢"1€ ™¢"1Ï¥"1L Ï¥"1Ó"1Ø Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1O:#1,O:#1ÍŽ#1˜ÍŽ#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÑ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UCƒÒ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÑ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UCƒÒ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÑ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UCƒÒ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ”´³ÓXx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ö6 €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1i é D Ä  Ÿ ú z ¦æ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Õ ÕJJ¿¿)4*44©4©> ? I•I•S T ^|^|hîhîr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÕÕJJ¿¿ 4 4 © ©““}}òòggÜÜQ Q"Æ"Æ$;%;'°'°)%*%,š,š./1„1„3ù3ù5n6n8ã8ã:X;X=Í=Í?B@BB·B·D,E,G¡G¡IJL�L�NOQwQwSìSìUaVaXÖXÖZK[K]À]À_5`5bªbªdeg”g”i j l~l~nðnðpbqbsÔsÔuFvFx¸x¸z*{*}œ}œ€‚€‚€„ò„ò†d‡d‰Ö‰Ö‹HŒHŽºŽº�,‘,“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1~!1Ð~!1Ÿ¾!1LŸ¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Óç!1(Óç!1q("1¤q("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1¥Q"1€¥Q"1C’"1üC’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1CÚ"1´ CÚ"1£þ"10 £þ"1å+#1På+#1:#1,:#1³�#1˜³�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1·€&1Ì·€&1‡R'1‡R'1/e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼Z& b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R’Ú²ò bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼Z& b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R’Ú²ò bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼Z&  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R’Ú²ò  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ³Ó‰©%E bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1À €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1µ 5 ‚  O Ï  œ �Ð €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙOOÅÅ);*;4±4±>'?'I›I›ST^ƒ^ƒh÷h÷rksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰ÿÿuuëëa a"×"×$M%M'Ã'Ã)9*9,¯,¯.%/%1›1›346‡6‡8ý8ý:s;s=é=é?_@_BÕBÕDKEKGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡‹‡‹‰ÿ‰ÿ‹sŒsŽçŽç�[‘[“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1}}!1Ð}}!1¾!1L¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Oç!1(Oç!1í'"1¤í'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1!Q"1€!Q"1¿‘"1ü¿‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#1©7#1,©7#1Ž#1˜Ž#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1ó€&1Ìó€&1ÃR'1ÃR'1ke'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÔ  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]E…à  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÔ  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]E…à  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÔ   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]E…à   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ³Ó~ž%E bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1JŠ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1à`¹9’ k ë Ú €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Þ ÞXXÎÎ)D*D4º4º>0?0I¤I¤ST^Œ^Œhistst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÞÞXXÒÒ L L Æ Æ@@ºº44®®((¢¢ – –"#%†%†'ü'ü)r*r,è,è.^/^1Ô1Ô3J4J6À6À8696;¬;¬=">"@˜@˜BCE„E„GúGúIpJpLäLäNXOXQÌQÌS@T@V´V´X(Y([œ[œ]^`„`„bøbødlelgàgàiTjTlÈlÈn/?/I¤I¤ST^Ž^Žhisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbébéd^e^gÓgÓiHjHl½l½n2o2q§q§stv‘v‘xy{{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1}}!1Ð}}!1¾!1L¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Oç!1(Oç!1í'"1¤í'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1!Q"1€!Q"1¿‘"1ü¿‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#1=:#1,=:#1�Ž#1˜�Ž#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÒž b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 \D„Þ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÒž b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 \D„Þ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÒž  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 \D„Þ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ³Ó—·^~ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ã# €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1D Ä  � ö v Ï O s³ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜTTÈÈ)<*<4°4°>$?$I–I–ST^z^zhìhìr^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœŒŒ | |"ô"ô$h%h'Ü'Ü)P*P,Ä,Ä.8/81¬1¬3 4 6”6”89;|;|=ð=ð?d@dBØBØDLELGÀGÀI4J4L¦L¦NOQŠQŠSüSüUnVnXàXàZR[R]Ä]Ä_6`6b¨b¨degŒgŒiþiþkplpnânâpTqTsÆsÆu8v8xªxªz{}Ž}Ž€‚r‚r„ä„ä†V‡V‰ȉȋ:Œ:ެެ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1û}!1Ðû}!1™¾!1L™¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Íç!1(Íç!1k("1¤k("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1ŸQ"1€ŸQ"1=’"1ü=’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#1:#1,:#1×�#1˜×�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±O b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]��Íð0 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±O b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]��Íð0 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±O  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]��Íð0  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ³ÓŠª+K bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1½ý €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ä D “  b â 1 ± ­í €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÅÅ):*:4¯4¯>$?$I—I—S T ^}^}hðhðrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=ß=ß?T@TBÉBÉD>E>G³G³I(J(L›L›NOQ�Q�SôSôUgVgXÚXÚZM[M]À]À_3`3b¦b¦degŒgŒiÿiÿkrlrnånåpXqXsËsËu>v>x±x±z${$}—}— € ‚}‚}„ð„ð†c‡c‰Ö‰Ö‹IŒI޼޼�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1ƒ}!1Ѓ}!1!¾!1L!¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1Uç!1(Uç!1ó'"1¤ó'"1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1'Q"1€'Q"1Å‘"1üÅ‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1[Ú"1´ [Ú"1»þ"10 »þ"1ý+#1Pý+#1—7#1,—7#1×�#1˜×�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÕ¡ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  I[›@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÕ¡ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  I[›@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÕ¡  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  I[›@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 }�0P'G bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations19y€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÃCšqñH È ©é €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×LLÁÁ)6*64«4«>?I‘I‘ST^w^whêhêr]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡üüqqææ[ ["Ð"Ð$E%E'º'º)/*/,¤,¤./1Ž1Ž346x6x8í8í:b;b=×=×?J@JB½B½D0E0G£G£IJL‰L‰NüNüPoQoSâSâUUVUXÈXÈZ;[;]®]®_!`!b”b”degzgziíiík`l`nÓnÓpFqFs¹s¹u,v,xŸxŸz{}…}…øø�k‚k„Þ„Þ†Q‡Q‰ÉË5Œ5ާާ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1ï}!1Ðï}!1�¾!1L�¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1Áç!1(Áç!1_("1¤_("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1‹7#1,‹7#1e�#1˜e�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1Û€&1ÌÛ€&1«R'1«R'1Se'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤@  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 \œ¦æâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤@  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 \œ¦æâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤@   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 \œ¦æâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 «Ëcƒf† bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1)i€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�`à1± ‚ 9y €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii ã ãYYÏÏ)E*E4»4»>/?/I£I£ST^‹^‹hÿhÿrsss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiãã]]×× Q Q Ë ËEE¿¿99³³))ŸŸ ‹ ‹"#%w%w'í'í)c*c,Ù,Ù.O/O1Å1Å3;4;6±6±8'9';�;�=>@‡@‡BûBûDoEoGãGãIWJWLËLËN?O?Q³Q³S'T'V›V›XY[ƒ[ƒ]÷]÷_k`kbßbßdSeSgÇgÇi;j;l¯l¯n#o#q—q—s t vvxóxózg{g}Û}ÛO€O‚ÂÄ7…7‡«‡«‰ŠŒ‘Œ‘Ž�‘w‘w“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1ƒ}!1Ѓ}!1!¾!1L!¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1Uç!1(Uç!1ó'"1¤ó'"1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1'Q"1€'Q"1Å‘"1üÅ‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#11:#1,1:#1mŽ#1˜mŽ#1™$1$™$1qò$1èqò$1«Þ%1ì«Þ%1€&1Ì€&1ëQ'1ëQ'1“d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÓŸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JZšþ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÓŸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JZšþ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÓŸ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JZšþ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 šº³Ód„ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ð €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 Ÿ ö v Í M ¤ $ @€ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙLL¿¿)2*24¥4¥>?I‡I‡SøSø]i^ihÚhÚrKsK} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰üüooââU U"È"È$;%;'®'®)!*!,”,”./1z1z3í3í5`6`8Ó8Ó:F;F=¹=¹?*@*B›B›D E G}G}IîIîK_L_NÐNÐPAQAS²S²U#V#X”X”Z[]v]v_ç_çaXbXdÉdÉf:g:i«i«kln�n�pþpþrosouàuàwQxQzÂzÂ|3}3¤¤�‚„†„††÷†÷ˆg‰g‹׋×�GŽG�·�·’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1õ}!1Ðõ}!1“¾!1L“¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Áç!1(Áç!1_("1¤_("1[/"1¬[/"1å8"1(å8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1:#1,:#1õ�#1˜õ�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1c€&1Ìc€&13R'13R'1Ûd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦D b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]�§çä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦D b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]�§çä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦D  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]�§çä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‹«Gg-M bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ºú €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ó S ¤ $ u õ F Æ Ê  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßSSÇÇ);*;4¯4¯>!?!I“I“ST^w^whéhér[s[} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ‡‡ûûo o"ã"ã$W%W'Ë'Ë)?*?,³,³.'/'1›1›346ƒ6ƒ8÷8÷:k;k=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b�b�dÿdÿfqgqiãiãkUlUnÇnÇp9q9s«s«uvx�x�z{}s}såå�W‚W„ɄɆ;‡;‰¬‰¬‹ŒŽŽŽŽ�ÿ�ÿ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1Çp7i!1‰}!1؉}!1'¾!1T'¾!1qÅ!1\qÅ!1ûÎ!1ØûÎ!11Ò!1¤1Ò!1[ç!10[ç!1ù'"1¬ù'"1C/"1´C/"1Í8"10Í8"1<"1ü<"1-Q"1ˆ-Q"1Ë‘"1Ë‘"1™"1 ™"1Ÿ¢"1ˆ Ÿ¢"1Õ¥"1T Õ¥"1Ó"1à Ó"1-Õ"1¼ -Õ"1¥Ø"1H ¥Ø"1ý"1´ ý"1G*#1ÔG*#1‘7#1°‘7#1¹�#1¹�#1½$1¨½$1•ò$1l•ò$1ÏÞ%1pÏÞ%1É€&1PÉ€&1™R'1”™R'1Ae'1˜€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ö¢ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 RCƒÉ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ö¢ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 RCƒÉ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ö¢  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 RCƒÉ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes1152patternContiguousinvocations1€Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1zš bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1*j€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¨(ÿVÖ- ­ ŠÊ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜQQÆÆ);*;4°4°>&?&IšIšST^‚^‚höhörlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœ††ûûp p"å"å$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ì=ì?a@aBÖBÖDKEKGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqvqvsêsêu^v^xÒxÒzF{F}º}º.€.‚¢‚¢„…‡Œ‡Œ‰ŠŒvŒvŽëŽë�`‘`“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1é}!1Ðé}!1‡¾!1L‡¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1»ç!1(»ç!1Y("1¤Y("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1�Q"1€�Q"1+’"1ü+’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1�Ú"1´ �Ú"1ýþ"10 ýþ"1?,#1P?,#1‘7#1,‘7#1¿�#1˜¿�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1/�&1Ì/�&1ÿR'1ÿR'1§e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™5 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 E…È‘Ñ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™5 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 E…È‘Ñ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™5  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 E…È‘Ñ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¤Äi‰Xx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1.n€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¸8� bâ7 · ŽÎ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii â â[[ÔÔ)M*M4Æ4Æ>???I¶I¶S+T+^ž^žhis„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiââ[[ÔÔ M M Æ Æ??¸¸11ªª##œœ Ž Ž"#%€%€'ù'ù)r*r,ë,ë.d/d1Ý1Ý3V4V6Ï6Ï8H9H;Á;Á=:>:@³@³B,C,E¥E¥GHJ—J—LMO…O…QüQüSsTsVêVêXaYa[Ø[Ø]O^O`Æ`Æb;c;e°e°g%h%jšjšlmo‚o‚qõqõshthvÛvÛxNyN{Á{Á}4~4€§€§‚ƒ…�…�‡ˆŠrŠrŒäŒäŽV�V‘ȑȓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1‰}!1Љ}!1'¾!1L'¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Uç!1(Uç!1ó'"1¤ó'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1'Q"1€'Q"1Å‘"1üÅ‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1%:#1,%:#1OŽ#1˜OŽ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ô  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 QCƒÊ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ô  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 QCƒÊ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ô   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 QCƒÊ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �½³Ód„ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Á €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 „ Û [ ² 2 ‰ !a €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94®4®>#?#I–I–ST^z^zhìhìr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?O@OBÄBÄD9E9G®G®I#J#L–L–N O Q|Q|SïSïUbVbXÕXÕZH[H]»]»_.`.b b deg„g„iöiökhlhnÚnÚpLqLs¾s¾u0v0x¢x¢z{}†}†øø�j‚j„܄܆P‡P‰ÉË6Œ6ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1ï}!1Ðï}!1�¾!1L�¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1»ç!1(»ç!1Y("1¤Y("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1�Q"1€�Q"1+’"1ü+’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1%:#1,%:#1UŽ#1˜UŽ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›9 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 D„Ç’Ò bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›9 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 D„Ç’Ò bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›9  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 D„Ç’Ò  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 „¤?d„ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1à €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 Œ á a ¶ 6 ‹ #c €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Þ ÞUUÌÌ)C*C4º4º>1?1I¦I¦ST^Š^Šhûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIYJYLÎLÎNCOCQ¸Q¸S-T-V¢V¢XY[Œ[Œ]^`v`vbébéd\e\gÏgÏiBjBlµlµn&o&q—q—stvyvyxêxêz[{[}Ì}Ì=€=‚®‚®„…‡‘‡‘‰ŠŒsŒsŽäŽä�U‘U“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1�}!1Ð�}!1-¾!1L-¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1aç!1(aç!1ÿ'"1¤ÿ'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"13Q"1€3Q"1Ñ‘"1üÑ‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1©7#1,©7#1…Ž#1˜…Ž#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ó€&1Ìó€&1ÃR'1ÃR'1ke'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|×£ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @D„Í  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|×£ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @D„Í  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|×£  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @D„Í   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 v–<Jj bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€âbÄD ¦ & P� €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>0?0I§I§ST^�^�histst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIXJXLÏLÏNFOFQ½Q½S4T4V«V«X"Y"[™[™]^`‡`‡bþbþdueugìgìicjclÚlÚnLoLq¾q¾s0t0v¢v¢xy{†{†}ø}øj€j‚܂܄N…N‡À‡À‰2Š2Œ¤Œ¤Ž�‘ˆ‘ˆ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1ã}!1Ðã}!1�¾!1L�¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1¯ç!1(¯ç!1M("1¤M("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1�Q"1€�Q"1’"1ü’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1¯7#1,¯7#1‹Ž#1˜‹Ž#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1G�&1ÌG�&1S'1S'1¿e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ*ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 HˆË �Ý bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ*ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 HˆË �Ý bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ*ö  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 HˆË �Ý  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ™¹d„Bb bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D„€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�ðpÐP ° 0 T” €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áYYÑÑ)I*I4Á4Á>9?9I¯I¯S%T%^›^›his‡s‡} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‘‘  � �"ù"ù$q%q'é'é)a*a,Ù,Ù.Q/Q1É1É3A4A6¹6¹8191;©;©=!>!@™@™BCE‰E‰GHJyJyLïLïNeOeQÛQÛSQTQVÇVÇX=Y=[³[³])^)`Ÿ`Ÿbce‹e‹ghjwjwlílíncocqÙqÙsOtOvÅvÅx;y;{±{±}'~'€�€�‚ƒ…‰…‰‡ÿ‡ÿ‰uŠuŒëŒëŽa�a‘בד b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1�}!1Ð�}!1-¾!1L-¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1[ç!1([ç!1ù'"1¤ù'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1-Q"1€-Q"1Ë‘"1üË‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1I:#1,I:#1'�#1˜'�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Ÿ€&1ÌŸ€&1oR'1oR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Õ¡ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @D„Î bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Õ¡ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @D„Î bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Õ¡  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @D„Î  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ³Ó–¶Zz bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1â" €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ˆ  j ê L Ì . ® ò2 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ù ÙNNÃÃ)8*84­4­>$?$I™I™ST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÙÙNNÃà 8 8 ­ ­""——  ��öökkààU U"Ê"Ê$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1ã}!1Ðã}!1�¾!1L�¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1µç!1(µç!1S("1¤S("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1‡Q"1€‡Q"1%’"1ü%’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1O:#1,O:#1'�#1˜'�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�.ú b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G‡Ê �Ý bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�.ú b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G‡Ê �Ý bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�.ú  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G‡Ê �Ý  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 y™ -cƒ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ä$ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�  p ð P Ð 0 ° ô4 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝSSÉÉ)?*?4µ4µ>+?+IŸIŸST^‡^‡hûhûroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��yyïïe e"Û"Û$Q%Q'Ç'Ç)=*=,³,³.)/)1Ÿ1Ÿ346‹6‹89;w;w=í=í?c@cBÙBÙDOEOGÅGÅI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÃbÃd7e7g«g«ijl“l“noq{q{sïsïucvcx×x×zK{K}¿}¿3€3‚§‚§„…‡�‡�‰ŠŒwŒwŽëŽë�_‘_“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1•}!1Е}!13¾!1L3¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1aç!1(aç!1ÿ'"1¤ÿ'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"13Q"1€3Q"1Ñ‘"1üÑ‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1Á7#1,Á7#1Q�#1˜Q�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ؤ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 GCƒÏ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ؤ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 GCƒÏ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ؤ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 GCƒÏ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 m� *Cc bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1W—€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1YÙFÆ3 ³   W €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh â â\\ÒÒ)H*H4¾4¾>4?4I¨I¨ST^�^�hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²,,¦¦  š š"#%Š%Š'(*v*v,ì,ì.b/b1Ø1Ø3N4N6Ä6Ä8:9:;°;°=&>&@œ@œBCEˆEˆGþGþItJtLèLèN\O\QÐQÐSDTDV¸V¸X,Y,[ [ ]^`ˆ`ˆbübüdpepgägäiXjXlÌlÌn@o@q´q´s(t(vœvœxy{„{„}ø}øl€l‚à‚à„T…T‡ȇȉ<Š<Œ°Œ°Ž$�$‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1Ý}!1ÐÝ}!1{¾!1L{¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1©ç!1(©ç!1G("1¤G("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1{Q"1€{Q"1’"1ü’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1Ç7#1,Ç7#1Q�#1˜Q�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒë b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹ÎœÜ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒë b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹ÎœÜ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒë  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹ÎœÜ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ž®]},L bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Zš€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hèSÓ> ¾ ) © Z €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii à àWWÎÎ)E*E4¼4¼>3?3I¨I¨ST^’^’his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiààWWÎÎ E E ¼ ¼33ªª!!˜˜††ýýt t"ë"ë$b%b'Ù'Ù)P*P,Ç,Ç.>/>1µ1µ3,4,6£6£89;‘;‘=>@@BöBöDmEmGäGäI[J[LÐLÐNEOEQºQºS/T/V¤V¤XY[Ž[Ž]^`x`xbíbídbebg×g×iLjLlÁlÁn6o6q«q«s t v•v•x y {{}ô}ôi€i‚Þ‚Þ„S…S‡ȇȉ=Š=Œ²Œ²Ž'�'‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1�}!1Ð�}!1-¾!1L-¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1aç!1(aç!1ÿ'"1¤ÿ'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"13Q"1€3Q"1Ñ‘"1üÑ‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1s:#1,s:#1ÿ�#1˜ÿ�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Ö¢ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 FCƒÐ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Ö¢ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 FCƒÐ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Ö¢  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 FCƒÐ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ³Ó�­Dd bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 � ú z ç g Ô TÄ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àXXÌÌ)@*@4´4´>(?(IšIšS T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  �� € €"ø"ø$l%l'à'à)T*T,È,È.Œ>ްް�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1Ý}!1ÐÝ}!1{¾!1L{¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1¯ç!1(¯ç!1M("1¤M("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1�Q"1€�Q"1’"1ü’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1�Ú"1´ �Ú"1ýþ"10 ýþ"1?,#1P?,#1s:#1,s:#1ÿ�#1˜ÿ�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1¥€&1Ì¥€&1uR'1uR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…#ï b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 JŠÍ �Ý bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…#ï b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 JŠÍ �Ý bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…#ï  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 JŠÍ �Ý  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 nŽûd„ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1E €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ” ÿ  ê j Õ UÅ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞTTÉÉ)>*>4³4³>(?(I›I›ST^�^�hôhôrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽzzððf f"Ü"Ü$Q%Q'Æ'Æ);*;,°,°.%/%1š1š346„6„8ù8ù:n;n=ã=ã?X@XBÍBÍDBEBG·G·I,J,LŸLŸNOQ…Q…SøSøUkVkXÞXÞZQ[Q]Ä]Ä_7`7bªbªdeg�g�ijlvlvnénép\q\sÏsÏuBvBxµxµz({(}›}›€‚�‚�„ô„ô†g‡g‰Ú‰Ú‹MŒMŽÀŽÀ�3‘3“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Å}!1�Å}!1c¾!1c¾!1³Â!1$³Â!1—ç!1�—ç!15("15("1…,"1$…,"1iQ"1�iQ"1’"1’"1W–"1$W–"1óØ"1�óØ"1Sý"1üSý"1/0#1 /0#1û~#1| û~#1Ï$1Ü Ï$1§ò$1  §ò$1áÞ%1¤ áÞ%1¥€&1„ ¥€&1uR'1ÈuR'1e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1)][�ê¶ Õ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1)][�ê¶ Õ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1)][�ê¶ Õ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ì ìaaÖÖ)K*K4¾4¾>1?1I¤I¤ST^Š^Šhýhýrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvììbbØØ N N Ä Ä::°°&&œœ††ûûp p"å"å$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ê=ê?]@]BÐBÐDCECG¶G¶I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnænæpYqYsÌsÌu?v?x²x²z%{%}˜}˜ € ‚~‚~„ñ„ñ†b‡b‰Ô‰Ô‹FŒFޏޏ�*‘*“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1×}!1�×}!1u¾!1 u¾!1_Å!1_Å!1©ç!1€©ç!1G("1üG("11/"11/"1{Q"1p{Q"1’"1ì’"1™"1ô™"1óØ"1`óØ"1Sý"1ÌSý"1/0#1ì/0#1û~#1L û~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1ù€&1T ù€&1ÉR'1˜ÉR'1qe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cwß b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1«ë€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cwß b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1«ë€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cwß b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1«ë €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ð ðffÜÜ)R*R4Æ4Æ>:?:I®I®S"T"^–^–h i s|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvððjjää ^ ^ Ø ØRRÌÌFFÀÀ66¬¬"" ˜ ˜"#%„%„'ú'ú)p*p,æ,æ.\/\1Ò1Ò3H4H6¾6¾8494;ª;ª=>@’@’BCEzEzGîGîIbJbLÖLÖNJOJQ¾Q¾S2T2V¦V¦XY[Ž[Ž]^`v`vbêbêd^e^gÒgÒiFjFlºlºn.o.q¢q¢stvŠvŠxþxþzr{r}æ}æZ€Z‚΂΄B…B‡´‡´‰'Š'ŒšŒšŽ � ‘€‘€“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ë}!1�Ë}!1i¾!1i¾!1³Â!1$³Â!1�ç!1��ç!1;("1;("1…,"1$…,"1oQ"1�oQ"1 ’"1 ’"1W–"1$W–"1óØ"1�óØ"1Sý"1üSý"1/0#1 /0#1û~#1| û~#1½$1Ü ½$1•ò$1  •ò$1ÏÞ%1¤ ÏÞ%1?€&1„ ?€&1R'1ÈR'1·d'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1*^Y�æ² Ø b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1*^Y�æ² Ø b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1*^Y�æ² Ø b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ì ì__ÒÒ)E*E4¶4¶>'?'I˜I˜S T ^z^zhëhërZsZ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvììbbØØ N N Ä Ä::°°&&œœ‚‚õõh h"Û"Û$N%N'Á'Á)4*4,§,§./1�1�346s6s8æ8æ:Y;Y=Ê=Ê?;@;B¬B¬DEGŽGŽIÿIÿKpLpNáNáPRQRSÃSÃU4V4X¥X¥Z[]‡]‡_ø_øaibidÚdÚfKgKi¼i¼k-l-nžnžpqs€s€uñuñwbxbzÓzÓ|D}Dµµ�&‚&„—„—†‡‰v‰v‹æ‹æ�VŽV�Æ�Æ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1×}!1�×}!1u¾!1 u¾!1_Å!1_Å!1©ç!1€©ç!1G("1üG("11/"11/"1{Q"1p{Q"1’"1ì’"1™"1ô™"1óØ"1`óØ"1Sý"1ÌSý"1/0#1ì/0#1û~#1L û~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cwá b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ªê€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cwá b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ªê€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Cwá b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ªê €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv î îbbÖÖ)J*J4¼4¼>.?.I I ST^„^„höhörfsf} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvîîffÞÞ V V Î ÎFF¾¾66®®""––  ~ ~"ò"ò$f%f'Ú'Ú)N*N,Â,Â.6/61ª1ª346’6’89;z;z=ì=ì?^@^BÐBÐDBEBG´G´I&J&L˜L˜N O Q|Q|SîSîU`V`XÒXÒZD[D]¶]¶_(`(bšbšd e g~g~iðiðkblbnÔnÔpFqFs¸s¸u*v*xœxœz{}€}€òò�d‚d„Ö„Ö†F‡F‰·‰·‹(Œ(ޙޙ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1§}!1�§}!1E¾!1 E¾!1³Â!1³Â!1yç!1€yç!1("1ü("1…,"1…,"1KQ"1pKQ"1é‘"1ìé‘"1W–"1ôW–"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1á€&1T á€&1±R'1˜±R'1Ye'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒç³ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1G€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒç³ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1G€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒç³ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1G €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu í íbb××)L*L4Á4Á>5?5I©I©ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuííeeÝÝ U U Í ÍEE½½55­­""——  � �"ö"ö$k%k'à'à)U*U,Ê,Ê.?/?1´1´3)4)6ž6ž89;ˆ;ˆ=ý=ý?r@rBçBçD\E\GÑGÑIEJEL¹L¹N-O-Q¡Q¡STV‰V‰XýXýZq[q]å]å_Y`YbÍbÍdAeAgµgµi)j)l�l�noq…q…sùsùumvmxáxázU{U}É}É=€=‚±‚±„%…%‡™‡™‰ŠŒƒŒƒŽøŽø�m‘m“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ý}!1�Ý}!1{¾!1 {¾!1MÅ!1MÅ!1©ç!1€©ç!1G("1üG("1/"1/"1{Q"1p{Q"1’"1ì’"1ñ˜"1ôñ˜"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1A�&1T A�&1S'1˜S'1¹e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€è b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1œÜ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€è b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1œÜ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€è b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1œÜ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ë ëddÝÝ)V*V4Ï4Ï>F?FI½I½S2T2^¥^¥his‰s‰} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrëëddÝÝ V V Ï ÏHHÁÁ::³³,,¥¥ — —"#%‰%‰'(*{*{,ô,ô.m/m1æ1æ3_4_6Ø6Ø8Q9Q;Ê;Ê=C>C@¼@¼B5C5E®E®G'H'JžJžLMOŒOŒQRTzTzVñVñXhYh[ß[ß]V^V`Í`ÍbBcBe·e·g,h,j¡j¡lmo‰o‰qüqüsotovâvâxUyU{È{È};~;€®€®‚!ƒ!…”…”‡ˆŠwŠwŒéŒéŽ[�[‘͓͑ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1¡}!1�¡}!1?¾!1 ?¾!1³Â!1³Â!1sç!1€sç!1("1ü("1…,"1…,"1EQ"1pEQ"1ã‘"1ìã‘"1W–"1ôW–"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠã¯ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1F€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠã¯ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1F€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠã¯ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1F €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss é é^^ÓÓ)H*H4½4½>0?0I£I£ST^‡^‡hùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðsséé__ÕÕ K K Á Á77­­##™™ƒƒøøm m"â"â$W%W'Ì'Ì)A*A,¶,¶.+/+1 1 346Š6Š8ÿ8ÿ:t;t=é=é?^@^BÓBÓDHEHG½G½I0J0L£L£NOQ‰Q‰SüSüUoVoXâXâZU[U]È]È_;`;b­b­deg‘g‘ijlulunçnçpYqYsËsËu=v=x¯x¯z!{!}“}“€‚w‚w„é„é†[‡[‰Ή΋AŒAŽ´Ž´�'‘'“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ý}!1�Ý}!1{¾!1 {¾!1MÅ!1MÅ!1¯ç!1€¯ç!1M("1üM("1/"1/"1�Q"1p�Q"1’"1ì’"1ñ˜"1ôñ˜"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1Ï€&1T Ï€&1ŸR'1˜ŸR'1Ge'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€ê b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ý€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€ê b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ý€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€ê b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ý €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr é é``××)N*N4Å4Å>:?:I¯I¯S"T"^“^“histst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrréé``×× N N Å Å<<³³**¡¡�� } }"ô"ô$k%k'â'â)Y*Y,Ð,Ð.G/G1¾1¾35456¬6¬8#9#;š;š=>@ˆ@ˆBÿBÿDvEvGíGíIbJbL×L×NLOLQÁQÁS6T6V«V«X Y [•[•] ^ ``bòbòdeeegØgØiKjKl¾l¾n/o/q q stv‚v‚xóxózd{d}Õ}ÕF€F‚·‚·„(…(‡˜‡˜‰ Š ŒzŒzŽëŽë�\‘\“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1¡}!1�¡}!1?¾!1 ?¾!1³Â!1³Â!1mç!1€mç!1 ("1ü ("1…,"1…,"1?Q"1p?Q"1Ý‘"1ìÝ‘"1W–"1ôW–"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ä° b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ä° b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ä° b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq è è__ÖÖ)M*M4Ä4Ä>;?;I²I²S)T)^›^›h i ss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqèè__ÖÖ M M Ä Ä;;²²))  ŽŽ | |"ó"ó$j%j'á'á)X*X,Ï,Ï.F/F1½1½34446«6«8"9";™;™=>@‡@‡BþBþDuEuGìGìIcJcLÚLÚNQOQQÈQÈS?T?V¶V¶X-Y-[¤[¤]^`’`’b c e€e€g÷g÷injnlålånWoWqÉqÉs;t;v­v­xy{‘{‘}~€u€u‚ç‚ç„Y…Y‡ˇˉ=Š=Œ¯Œ¯Ž!�!‘“‘““ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1ã}!1�ã}!1�¾!1 �¾!1YÅ!1YÅ!1¯ç!1€¯ç!1M("1üM("1+/"1+/"1�Q"1p�Q"1’"1ì’"1ý˜"1ôý˜"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1M�&1T M�&1S'1˜S'1Åe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰%ñ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1±ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰%ñ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1±ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰%ñ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1±ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn æ æ^^ÖÖ)N*N4Æ4Æ>>¶¶..¦¦–– † †"þ"þ$v%v'î'î)f*f,Þ,Þ.V/V1Î1Î3F4F6¾6¾8696;®;®=&>&@ž@žBCEŽEŽGHJ|J|LòLòNhOhQÞQÞSTTTVÊVÊX@Y@[¶[¶],^,`¢`¢bceŽeŽghjzjzlðlðnfofqÜqÜsRtRvÈvÈx>y>{´{´}*~*€ € ‚ƒ…Œ…Œ‡ˆŠxŠxŒîŒîŽd�d‘Ú‘Ú“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1›}!1�›}!19¾!1 9¾!1³Â!1³Â!1mç!1€mç!1 ("1ü ("1…,"1…,"1?Q"1p?Q"1Ý‘"1ìÝ‘"1W–"1ôW–"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡à¬ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡à¬ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡à¬ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo ä äYYÎÎ)C*C4¸4¸>-?-I¢I¢ST^‰^‰hûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooääYYÎÎ C C ¸ ¸--¢¢ŒŒvvëë` `"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%LšLšNOQ„Q„SùSùUnVnXãXãZX[X]Í]Í_B`Bb·b·d,e,g¡g¡ijl‹l‹nýnýpoqosásáuSvSxÅxÅz7{7}©}©€‚�‚�„ÿ„ÿ†q‡q‰ã‰ã‹UŒUŽÇŽÇ�9‘9“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1ã}!1�ã}!1�¾!1 �¾!1YÅ!1YÅ!1µç!1€µç!1S("1üS("1+/"1+/"1‡Q"1p‡Q"1%’"1ì%’"1ý˜"1ôý˜"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Õ€&1T Õ€&1¥R'1˜¥R'1Me'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰'ó b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1°ð€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰'ó b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1°ð€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰'ó b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1°ð €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn ä äZZÐÐ)F*F4¼4¼>0?0I¤I¤ST^Œ^Œhistst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnääZZÐÐ F F ¼ ¼22¨¨””  €€ööl l"â"â$X%X'Î'Î)D*D,º,º.0/01¦1¦346’6’89;~;~=ô=ô?j@jBàBàDVEVGÌGÌI@J@L´L´N(O(QœQœSTV„V„XøXøZl[l]à]à_T`TbÈbÈd;?;I¯I¯S#T#^—^—h i ss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqëëeeßß Y Y Ó ÓMMÇÇAA»»55¯¯)) £ £"#%“%“' ( **,õ,õ.k/k1á1á3W4W6Í6Í8C9C;¹;¹=/>/@¥@¥BCE‘E‘GHJ{J{LïLïNcOcQ×Q×SKTKV¿V¿X3Y3[§[§]^`�`�bcewewgëgëi_j_lÓlÓnGoGq»q»s/t/v£v£xy{‹{‹}ÿ}ÿs€s‚ç‚ç„[…[‡χωCŠCŒ·Œ·Ž+�+‘Ÿ‘Ÿ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1ã}!1�ã}!1�¾!1 �¾!15Å!15Å!1µç!1€µç!1S("1üS("1/"1/"1‡Q"1p‡Q"1%’"1ì%’"1Ù˜"1ôÙ˜"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1 �&1T  �&1ÛR'1˜ÛR'1ƒe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’.ú b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Í€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’.ú b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Í€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’.ú b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Í €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj á áXXÏÏ)F*F4½4½>2?2I§I§ST^‘^‘his{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjááXXÏÏ F F ½ ½44««""™™‡‡þþu u"ì"ì$c%c'Ú'Ú)Q*Q,È,È.?/?1¶1¶3-4-6¤6¤89;’;’= > @€@€B÷B÷DnEnGåGåIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbìbìdaeagÖgÖiKjKlÀlÀn5o5qªqªstv”v”x y {~{~}ó}óh€h‚݂݄R…R‡LJlj<Š<Œ±Œ±Ž&�&‘›‘›“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1•}!1�•}!13¾!1 3¾!1ÅÂ!1ÅÂ!1gç!1€gç!1("1ü("1—,"1—,"19Q"1p9Q"1ב"1ìב"1i–"1ôi–"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„Ý© b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1_€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„Ý© b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1_€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„Ý© b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1_ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo ç ç__ÓÓ)G*G4»4»>-?-IŸIŸST^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooçç__×× O O Ç Ç??··//§§—— ‡ ‡"ÿ"ÿ$s%s'ç'ç)[*[,Ï,Ï.C/C1·1·3+4+6Ÿ6Ÿ89;‡;‡=û=û?o@oBãBãDWEWGËGËI=J=L¯L¯N!O!Q“Q“STVwVwXéXéZ[[[]Í]Í_?`?b±b±d#e#g•g•ijlylynënëp]q]sÏsÏuAvAx³x³z%{%}—}— € ‚{‚{„í„í†_‡_‰щыCŒC޵޵�'‘'“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1é}!1�é}!1‡¾!1 ‡¾!15Å!15Å!1µç!1€µç!1S("1üS("1/"1/"1‡Q"1p‡Q"1%’"1ì%’"1Ù˜"1ôÙ˜"1çØ"1`çØ"1Gý"1ÌGý"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’0ü b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Í€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’0ü b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Í€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’0ü b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Í €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³Ó  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk á áWWÌÌ)A*A4¶4¶>)?)IœIœST^‚^‚hõhõrhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkááWWÍÍ C C ¹ ¹//¥¥‘‘}}óói i"ß"ß$T%T'É'É)>*>,³,³.(/(1�1�346‡6‡8ü8ü:q;q=æ=æ?[@[BÐBÐDEEEGºGºI-J-L L NOQ†Q†SùSùUlVlXßXßZR[R]Å]Å_8`8b«b«deg‘g‘ijlwlwnênêp]q]sÐsÐuCvCx¶x¶z){)}œ}œ€‚‚‚‚„õ„õ†h‡h‰Û‰Û‹NŒNŽÁŽÁ�4‘4“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1�}!1��}!1-¾!1 -¾!1³Â!1³Â!1aç!1€aç!1ÿ'"1üÿ'"1…,"1…,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1W–"1ôW–"1UÚ"1`UÚ"1µþ"1̵þ"10#1ì0#1é~#1L é~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OƒÞª b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OƒÞª b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OƒÞª b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1%E  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj à àUUÊÊ)?*?4²4²>%?%I˜I˜S T ^~^~hñhñrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjààVVÌÌ B B ¸ ¸..¤¤��zzïïd d"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=Þ=Þ?Q@QBÄBÄD7E7GªGªIJL�L�NOQvQvSéSéU\V\XÏXÏZB[B]µ]µ_(`(b›b›deg�g�iôiôkglgnÚnÚpMqMsÀsÀu3v3x¦x¦z{}Œ}Œÿÿ�r‚r„å„å†V‡V‰ȉȋ:Œ:ެެ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1é}!1�é}!1‡¾!1 ‡¾!1SÅ!1SÅ!1»ç!1€»ç!1Y("1üY("1%/"1%/"1�Q"1p�Q"1+’"1ì+’"1÷˜"1ô÷˜"1sÚ"1`sÚ"1Óþ"1ÌÓþ"10#1ì0#1é~#1L é~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›7 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1®î€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›7 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1®î€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›7 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1®î €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Aa  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ä äZZÐÐ)F*F4º4º>.?.I¢I¢ST^Š^Šhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjää^^ØØ R R Ì ÌFFÀÀ::´´**   Œ Œ"#%x%x'î'î)d*d,Ú,Ú.P/P1Æ1Æ3<4<6²6²8(9(;ž;ž=>@†@†BúBúDnEnGâGâIVJVLÊLÊN>O>Q²Q²S&T&VšVšXY[‚[‚]ö]ö_j`jbÞbÞdReRgÆgÆi:j:l®l®n"o"q–q–s t v~v~xòxòzf{f}Ú}ÚN€N‚‚„6…6‡¨‡¨‰ŠŒŽŒŽŽ�‘t‘t“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1�}!1��}!1-¾!1 -¾!1³Â!1³Â!1[ç!1€[ç!1ù'"1üù'"1…,"1…,"1-Q"1p-Q"1Ë‘"1ìË‘"1W–"1ôW–"1gÚ"1`gÚ"1Çþ"1ÌÇþ"10#1ì0#1é~#1L é~#1Ÿ$1¬ Ÿ$1wò$1p wò$1±Þ%1t ±Þ%1!€&1T !€&1ñQ'1˜ñQ'1™d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Ú¦ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Ú¦ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Ú¦ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations18X  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj à àSSÆÆ)9*94ª4ª>?IŒIŒSýSý]n^nhßhßrNsN} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjààVVÌÌ B B ¸ ¸..¤¤��vvéé\ \"Ï"Ï$B%B'µ'µ)(*(,›,›./1�1�3ô3ô5g6g8Ú8Ú:M;M=¾=¾?/@/B B DEG‚G‚IóIóKdLdNÕNÕPFQFS·S·U(V(X™X™Z [ ]{]{_ì_ìa]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ï}!1�ï}!1�¾!1 �¾!1MÅ!1MÅ!1»ç!1€»ç!1Y("1üY("1/"1/"1�Q"1p�Q"1+’"1ì+’"1ñ˜"1ôñ˜"1‹Ú"1`‹Ú"1ëþ"1Ìëþ"10#1ì0#1é~#1L é~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1i€&1T i€&19R'1˜9R'1ád'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›9 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1­í€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›9 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1­í€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›9 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1­í €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Xx  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âVVÊÊ)>*>4°4°>"?"I”I”ST^x^xhêhêrZsZ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢ŠŠþþr r"æ"æ$Z%Z'Î'Î)B*B,¶,¶.*/*1ž1ž346†6†8ú8ú:n;n=à=à?R@RBÄBÄD6E6G¨G¨IJLŒLŒNþNþPpQpSâSâUTVTXÆXÆZ8[8]ª]ª_`bŽbŽdegrgriäiäkVlVnÈnÈp:q:s¬s¬uvx�x�z{}t}tææ�X‚X„ʄʆ:‡:‰«‰«‹ŒŽ�Ž��þ�þ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1‰}!1�‰}!1'¾!1 '¾!1¹Â!1¹Â!1[ç!1€[ç!1ù'"1üù'"1‹,"1‹,"1-Q"1p-Q"1Ë‘"1ìË‘"1]–"1ô]–"1[Ú"1`[Ú"1»þ"1Ì»þ"10#1ì0#1é~#1L é~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Û§ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1[€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Û§ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1[€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Û§ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1[ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1,L  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áVVËË)@*@4µ4µ>)?)I�I�ST^…^…hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‹‹ u u"ê"ê$_%_'Ô'Ô)I*I,¾,¾.3/31¨1¨346’6’89;|;|=ñ=ñ?f@fBÛBÛDPEPGÅGÅI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒwŒwŽìŽì�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ï}!1�ï}!1�¾!1 �¾!1)Å!1)Å!1Áç!1€Áç!1_("1ü_("1û."1û."1“Q"1p“Q"11’"1ì1’"1͘"1ô͘"1sÚ"1`sÚ"1Óþ"1ÌÓþ"10#1ì0#1é~#1L é~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1)�&1T )�&1ùR'1˜ùR'1¡e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤@  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Í€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤@  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Í€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤@  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Í €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Hh  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff ß ßXXÑÑ)J*J4Ã4Ã>:?:I±I±S&T&^™^™h i s}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffßßXXÑÑ J J à Ã<<µµ..§§  ™™ ‹ ‹"#%}%}'ö'ö)o*o,è,è.a/a1Ú1Ú3S4S6Ì6Ì8E9E;¾;¾=7>7@°@°B)C)E¢E¢GHJ’J’L M O€O€Q÷Q÷SnTnVåVåX\Y\[Ó[Ó]J^J`Á`Áb6c6e«e«g h j•j•l m o}o}qðqðsctcvÖvÖxIyI{¼{¼}/~/€¢€¢‚ƒ…ˆ…ˆ‡ù‡ù‰kŠkŒ݌ݎO�O‘Á‘Á“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1‰}!1�‰}!1'¾!1 '¾!1¹Â!1¹Â!1Uç!1€Uç!1ó'"1üó'"1‹,"1‹,"1'Q"1p'Q"1Å‘"1ìÅ‘"1]–"1ô]–"1sÚ"1`sÚ"1Óþ"1ÌÓþ"10#1ì0#1é~#1L é~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~×£ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1\€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~×£ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1\€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~×£ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1\ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Cc  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝRRÇÇ)<*<4±4±>$?$I—I—S T ^{^{híhír_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��wwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD.?.I£I£ST^‡^‡høhørhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÝÝTTËË B B ¹ ¹00§§••  ƒƒúúq q"è"è$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbæbædYeYgÌgÌi?j?l²l²n#o#q”q”stvvvvxçxçzX{X}É}É:€:‚«‚«„…‡Œ‡Œ‰ý‰ý‹nŒnŽߎß�P‘P“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1ƒ}!1�ƒ}!1!¾!1 !¾!1§Â!1§Â!1Uç!1€Uç!1ó'"1üó'"1y,"1y,"1'Q"1p'Q"1Å‘"1ìÅ‘"1K–"1ôK–"1[Ú"1`[Ú"1»þ"1Ì»þ"10#1ì0#1ã~#1L ã~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1í€&1T í€&1½R'1˜½R'1ee'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ؤ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ؤ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ؤ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations11Q  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>/?/I¦I¦ST^‘^‘hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÎLÎNEOEQ¼Q¼S3T3VªVªX!Y![˜[˜]^`†`†býbýdtetgëgëibjblÙlÙnMoMq¿q¿s1t1v£v£xy{‡{‡}ù}ùk€k‚݂݄O…O‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1õ}!1�õ}!1“¾!1 “¾!1GÅ!1GÅ!1Çç!1€Çç!1e("1üe("1/"1/"1™Q"1p™Q"17’"1ì7’"1ë˜"1ôë˜"1UÚ"1`UÚ"1µþ"1̵þ"10#1ì0#1ã~#1L ã~#1Õ$1¬ Õ$1ýð$1p ýð$1ÛÀ%1t ÛÀ%1;�&1T ;�&11Q'1˜1Q'1Ùc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­I b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1«ë€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­I b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1«ë€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­I b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1«ë €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1MI  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚRRÊÊ)B*B4º4º>0?0I¦I¦ST^œ^œhisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""ššŠŠ z z"ò"ò$j%j'â'â)Z*Z,Ò,Ò.J/J1Â1Â3:4:6²6²8*9*;¢;¢=>@’@’B C E‚E‚GúGúIpJpLæLæN\O\QÒQÒSHTHV¾V¾X4Y4[ª[ª] ^ `–`–b c e‚e‚gøgøinjnläländodqÚqÚsPtPvÆvÆx!?!I–I–S T ^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccØØMM 7 7 ¬ ¬!!––  €€õõjjßßT T"É"É$>%>'³'³)(*(,�,�./1‡1‡3ü3ü5q6q8æ8æ:[;[=Ð=Ð?E@EBºBºD/E/G¤G¤IJLŽLŽNOQxQxSíSíUbVbX×X×ZL[L]Á]Á_6`6b«b«d e g•g•i j llnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1õ}!1�õ}!1“¾!1 “¾!1GÅ!1GÅ!1Çç!1€Çç!1e("1üe("1/"1/"1™Q"1p™Q"17’"1ì7’"1ë˜"1ôë˜"1sÚ"1`sÚ"1Óþ"1ÌÓþ"10#1ì0#1ã~#1L ã~#1½$1¬ ½$1åð$1p åð$1ÃÀ%1t ÃÀ%1À&1T À&1¹P'1˜¹P'1ac'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­K b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ªê€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­K b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ªê€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­K b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ªê €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1pl  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÄÄ):*:4°4°>$?$I˜I˜S T ^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$L%L'Â'Â)8*8,®,®.$/$1š1š346†6†8ü8ü:r;r=è=è?^@^BÔBÔDJEJGÀGÀI4J4L¨L¨NOQ�Q�STVxVxXìXìZ`[`]Ô]Ô_H`Hb¼b¼d0e0g¤g¤ijlŒlŒnoq|q|sðsðudvdxØxØzL{L}À}À4€4‚¨‚¨„…‡�‡�‰ŠŒxŒxŽìŽì�`‘`“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1}}!1�}}!1¾!1 ¾!1­Â!1­Â!1Oç!1€Oç!1í'"1üí'"1,"1,"1!Q"1p!Q"1¿‘"1ì¿‘"1Q–"1ôQ–"1·Ø"1`·Ø"1ý"1Ìý"10#1ì0#1Ý~#1L Ý~#1É$1¬ É$1ñð$1p ñð$1ÏÀ%1t ÏÀ%1ÿ€&1T ÿ€&1õP'1˜õP'1�c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÕ¡ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1X€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÕ¡ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1X€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÕ¡ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1X €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³¯  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee ß ßYYÑÑ)G*G4½4½>1?1I¥I¥ST^™^™h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯))££ — —"#%‰%‰'ÿ'ÿ)u*u,ë,ë.a/a1×1×3M4M6Ã6Ã8999;¯;¯=%>%@›@›BCE‡E‡GýGýIqJqLåLåNYOYQÍQÍSATAVµVµX)Y)[�[�]^`…`…bùbùdmemgágáiUjUlÉlÉnIoIq½q½s1t1v¥v¥xy{�{�}~€u€u‚é‚é„]…]‡чщEŠEŒ¹Œ¹Ž-�-‘¡‘¡“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1û}!1�û}!1™¾!1 ™¾!1#Å!1#Å!1Çç!1€Çç!1e("1üe("1õ."1õ."1™Q"1p™Q"17’"1ì7’"1ǘ"1ôǘ"1·Ø"1`·Ø"1ý"1Ìý"10#1ì0#1Ý~#1L Ý~#1É$1¬ É$1ñð$1p ñð$1ÏÀ%1t ÏÀ%1�&1T �&1ûP'1˜ûP'1£c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶R b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Í€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶R b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Í€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶R b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Í €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³¯  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ Õ ÕLLÍÍ)D*D4»4»>0?0I¥I¥ST^›^›his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÕÕLLÃà : : ± ±((ŸŸ��{{òòi i"à"à$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubêbêd_e_gÔgÔiIjIl¾l¾n?o?q´q´s)t)vžvžxy{ˆ{ˆ}ý}ýr€r‚ç‚ç„\…\‡чщFŠFŒ»Œ»Ž0�0‘¥‘¥“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1w}!1�w}!1¾!1 ¾!1­Â!1­Â!1Iç!1€Iç!1ç'"1üç'"1,"1,"1Q"1pQ"1¹‘"1칑"1Q–"1ôQ–"1·Ø"1`·Ø"1ý"1Ìý"10#1ì0#1Ý~#1L Ý~#1±$1¬ ±$1Ùð$1p Ùð$1·À%1t ·À%1‡€&1T ‡€&1}P'1˜}P'1%c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÑ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Y€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÑ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Y€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1DxÑ� b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Y €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³¯  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSËË)?*?4³4³>%?%I—I—S T ^‡^‡hùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnïnïpaqasÓsÓuEvEx·x·z){)}›}› € ‚‚„ñ„ñ†c‡c‰Õ‰Õ‹GŒG޹޹�+‘+“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1û}!1�û}!1™¾!1 ™¾!1#Å!1#Å!1Íç!1€Íç!1k("1ük("1õ."1õ."1ŸQ"1pŸQ"1=’"1ì=’"1ǘ"1ôǘ"1·Ø"1`·Ø"1ý"1Ìý"10#1ì0#1Ý~#1L Ý~#1·$1¬ ·$1ßð$1p ßð$1½À%1t ½À%1™€&1T ™€&1�P'1˜�P'17c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶T  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Í€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶T  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Í€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶T  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Í €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1³¯  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Õ ÕKKÈÈ)=*=4²4²>%?%I˜I˜S T ^Š^Šhýhýrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……ûûqqçç] ]"Ó"Ó$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnònòpeqesØsØuKvKx¾x¾z1{1}¤}¤€‚ЂЄý„ý†p‡p‰ã‰ã‹VŒVŽÉŽÉ�<‘<“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„%i!1G}!1G}!1å½!1|å½!1IÃ!1„IÃ!1Æ!1ðÆ!1 Ð!1| Ð!1?Ó!1H?Ó!1ç!1Ôç!1·'"1P·'"1-"1X-"1Q0"1ÄQ0"1Û9"1PÛ9"1="1="1ëP"1¨ëP"1‰‘"1$ ‰‘"1í–"1, í–"1#š"1˜ #š"1­£"1$ ­£"1ã¦"1ð ã¦"1%Ô"1| %Ô"1[Ú"1X[Ú"1»þ"1Ô»þ"1ý+#1ôý+#1+:#1Ð+:#1çn#1\çn#1G�#1€G�#1E˜#1¬E˜#1É$18É$1¡ò$1ü¡ò$1ÛÞ%1ÛÞ%1Uì%1àUì%1ÿ€&1lÿ€&1ÏR'1°ÏR'1we'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P­y b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1µõ bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P­y b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1µõ  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P­y  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1µõ bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 šºµÕDd bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1E…#cÊ O�Ô  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dEÅ#£ ŠòrÏOÁA”�YÙžãc¨(mí3³øx½= ‚  G Ç  Œ Ñ Q – Ö  [  S   à %e bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ ­ ­LLÍÍ)D*D4»4»>2?2I§I§ST^�^�his‡s‡} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÕÕLLÃà : : ‰ ‰wwîîee{{òòi i"à"à$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbìbìdaeagÖgÖiKjKlÀlÀnAoAq¶q¶s+t+v v xy{Š{Š}ÿ}ÿt€t‚é‚é„^…^‡Ó‡Ó‰HŠHŒ½Œ½Ž2�2‘§‘§“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1±¾!1d±¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1ßç!1@ßç!1}("1¼}("11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1±Q"1˜±Q"1O’"1O’"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"17Ú"1Ì 7Ú"1—þ"1H —þ"1Ù+#1hÙ+#1:#1D:#1;o#1Ð;o#1ëŽ#1ôëŽ#1é™#1 é™#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1À&1àÀ&1“R'1$“R'1;e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒp< b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 E…•ÕÊ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒp< b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 E…•ÕÊ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒp<  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 E…•ÕÊ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 µÕ†¦< bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1%eY™ò2�Í(h  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d%¥YÙîn4´·7 �€æf¿?˜qñJÊ%¥€Û[ ¶ 6 ‘  l ì G Ç µ õ " ¢ û ;”Ô/oÊ  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛPPÇÇ)*4±4±>'?'I›I›ST^�^�hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››……úúo o"ä"ä$[%['Ð'Ð)E*E,º,º.///1|1|3ñ3ñ5f6f8Û8Û:P;P=í=í?b@bB×B×DLELGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`KbËbËd?e?g³g³i'j'l›l›noqƒqƒs÷s÷ukvkxßxßzS{S}Ç}Ç;€;‚¯‚¯„#…#‡™‡™‰ŠŒƒŒƒŽøŽø�m‘m“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„%i!1G}!1G}!1å½!1|å½!1IÃ!1„IÃ!1Æ!1ðÆ!1 Ð!1| Ð!1?Ó!1H?Ó!1ç!1Ôç!1±'"1P±'"1-"1X-"1Q0"1ÄQ0"1Û9"1PÛ9"1="1="1åP"1¨åP"1ƒ‘"1$ ƒ‘"1í–"1, í–"1#š"1˜ #š"1­£"1$ ­£"1ã¦"1ð ã¦"1%Ô"1| %Ô"1OÚ"1XOÚ"1¯þ"1Ô¯þ"1ñ+#1ôñ+#1+:#1Ð+:#1çn#1\çn#1G�#1€G�#1E˜#1¬E˜#1±$18±$1‰ò$1ü‰ò$1ÃÞ%1ÃÞ%1=ì%1à=ì%1“€&1l“€&1cR'1°cR'1 e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oªv b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1µõ bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oªv b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1µõ  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oªv  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1µõ bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 zš3S9 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1E…Ê ²òO�Ô  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dEÅ ŠÏO”YÙ_ßž.®ãcý}¨(ÌLmí2²÷w¼< �  F Æ  ‹ Ð P • Õ  Z Ÿ ß ¢ â $d bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ ¬ ¬KKÈÈ)=*=4²4²>'?'IšIšS T ^Œ^Œhÿhÿrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 „ „úúppææ\\ûûqqçç] ]"Ó"Ó$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I+J+LžLžNOQ„Q„S÷S÷UjVjXÝXÝZP[P]Ã]Ã_6`6b©b©deg�g�ijlulunônôpgqgsÚsÚuMvMxÀxÀz3{3}¦}¦€‚Œ‚Œ„ÿ„ÿ†r‡r‰å‰å‹XŒXŽËŽË�>‘>“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1±¾!1d±¾!1_Å!1l_Å!1éÎ!1èéÎ!1Ò!1´Ò!1åç!1@åç!1ƒ("1¼ƒ("11/"1Ä1/"1»8"1@»8"1ñ;"1 ñ;"1·Q"1˜·Q"1U’"1U’"1™"1 ™"1�¢"1˜ �¢"1Ã¥"1d Ã¥"1Ó"1ð Ó"1=Ú"1Ì =Ú"1�þ"1H �þ"1ß+#1hß+#1Á:#1DÁ:#1}o#1Ð}o#1A�#1ôA�#1?›#1 ?›#1«$1¬«$1ƒò$1pƒò$1½Þ%1t½Þ%17ì%1T7ì%1�€&1à�€&1QR'1$QR'1ùd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ôt@ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 E…–ÖÊ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ôt@ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 E…–ÖÊ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ôt@  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 E…–ÖÊ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 µÕy™"B bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1l¬®îXÂl¬  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dlìXØDÄ0°IÉœ’ˆÛ[¤$ôtà`ÊJ´4 ž  ˆ  r ò \ Ü F Æ > ~ 0°\Èr²\ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×LLÅÅ)*4¯4¯>$?$I—I—ST^‡^‡hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡üüqqææ[ ["Ð"Ð$I%I'¾'¾)3*3,¨,¨./1i1i3Þ3Þ5S6S8È8È:=;==Û=Û?P@PBÅBÅD:E:G¯G¯I$J$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b­b­deg‘g‘ijlulunçnçpYqYsËsËu=v=x¯x¯z!{!}“}“€‚w‚w„é„é†]‡]‰ЉЋCŒC޶޶�)‘)“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1}}!1è}}!1¾!1d¾!1eÅ!1leÅ!1ïÎ!1èïÎ!1%Ò!1´%Ò!1Iç!1@Iç!1ç'"1¼ç'"17/"1Ä7/"1Á8"1@Á8"1÷;"1 ÷;"1Q"1˜Q"1¹‘"1¹‘"1 ™"1  ™"1“¢"1˜ “¢"1É¥"1d É¥"1 Ó"1ð  Ó"1aÚ"1Ì aÚ"1Áþ"1H Áþ"1,#1h,#1µ:#1Dµ:#1qo#1Ðqo#1á�#1ôá�#1ßš#1 ßš#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1Õ€&1àÕ€&1¥R'1$¥R'1Me'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÖ¢ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 $dE…Ê  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÖ¢ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 $dE…Ê  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÖ¢  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 $dE…Ê   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ™¹µÕBb bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1 `f¦ K²òY™  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d  fæînKË»;0°ˆ•úzß_ÄD©)�w÷^ Þ E Å , ¬  “ ú z ñ 1 á aÆk«R¹ù bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ü ÜRRÈÈ)>*>4´4´>?IœIœST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÜÜVVÐÐ J J Ä Ä>>¸¸22¬¬""˜˜ „ „"ú"ú$p%p'æ'æ)\*\,Ò,Ò.H/H1¾1¾34446ª6ª8 9 ;–;–= > @€@€BôBôDhEhGÜGÜI(J(LœLœNOQ„Q„SøSøU”V”XY[|[|]ð]ð_d`dbâbâdVeVgÊgÊi>j>l²l²n&o&qšqšstv‚v‚xöxözj{j}Þ}ÞR€R‚ƂƄ:…:‡®‡®‰!Š!Œ”Œ”Ž�‘z‘z“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1~!1è~!1¥¾!1d¥¾!1eÅ!1leÅ!1ïÎ!1èïÎ!1%Ò!1´%Ò!1Ùç!1@Ùç!1w("1¼w("17/"1Ä7/"1Á8"1@Á8"1÷;"1 ÷;"1«Q"1˜«Q"1I’"1I’"1 ™"1  ™"1“¢"1˜ “¢"1É¥"1d É¥"1 Ó"1ð  Ó"1=Ú"1Ì =Ú"1�þ"1H �þ"1ß+#1hß+#1a:#1Da:#1o#1Ðo#1=Ž#1ô=Ž#1;™#1 ;™#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1á€&1àá€&1±R'1$±R'1Ye'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Çe1 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 E…—×Ê  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Çe1 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 E…—×Ê  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Çe1  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 E…—×Ê   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 µÕ‡§> bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1 `P�à r²D  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d  PÐëk  ¶6ðp�À@�`à0°€ÒR¤$vöH È  š ì l ¾ > �  Ó  `   ð 0‚ÂT bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>/?/I¦I¦ST^‘^‘hÛhÛrusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIWJWLÎLÎNEOEQ¼Q¼S3T3VªVªX!Y![˜[˜]^`†`†býbýdtetgëgëibjblÙlÙnMoMq¿q¿s1t1v£v£xy{_{_}Ñ}ÑC€C‚µ‚µ„'…'‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1w}!1èw}!1¾!1d¾!1eÅ!1leÅ!1ïÎ!1èïÎ!1%Ò!1´%Ò!1Iç!1@Iç!1ç'"1¼ç'"17/"1Ä7/"1Á8"1@Á8"1÷;"1 ÷;"1Q"1˜Q"1¹‘"1¹‘"1 ™"1  ™"1“¢"1˜ “¢"1É¥"1d É¥"1 Ó"1ð  Ó"17Ú"1Ì 7Ú"1—þ"1H —þ"1Ù+#1hÙ+#1¯:#1D¯:#1ko#1Ðko#1É�#1ôÉ�#1Çš#1 Çš#1Ÿ$1¬Ÿ$1wò$1pwò$1±Þ%1t±Þ%1+ì%1T+ì%1]€&1à]€&1-R'1$-R'1Õd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÔ  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 #cE…Ê  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÔ  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 #cE…Ê  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1EyÔ   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 #cE…Ê   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 y™µÕ; bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1f¦­í L°ðT”  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dfæLÌ2²˜UÕþ~"¢ädïoÊJ¼<°0”xø\ Ü @ À $ ¤  ˆ ì l ~ ¾ Ð P¶ö\œ@¤ä bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚNNÂÂ)6*64ª4ª>ó>óHŽIŽST^z^zhìhìr^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""šš‚‚ööj j"Þ"Þ$R%R'Æ'Æ):*:,®,®."/"1–1–3 4 6~6~8ò8ò:f;f=Ú=Ú?L@LB¾B¾D0E0G¢G¢IëIëK]L]NÏNÏPAQAS³S³UNVNXÀXÀZ2[2]¤]¤_`b�b�degtgtiæiækXlXnÊnÊp‡>‰¯‰¯‹ Œ ޑޑ�‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1 ~!1è ~!1«¾!1d«¾!1eÅ!1leÅ!1ïÎ!1èïÎ!1%Ò!1´%Ò!1ßç!1@ßç!1}("1¼}("17/"1Ä7/"1Á8"1@Á8"1÷;"1 ÷;"1±Q"1˜±Q"1O’"1O’"1 ™"1  ™"1“¢"1˜ “¢"1É¥"1d É¥"1 Ó"1ð  Ó"1IÚ"1Ì IÚ"1©þ"1H ©þ"1ë+#1hë+#11:#1D1:#1ín#1Ðín#1M�#1ôM�#1K˜#1 K˜#1«$1¬«$1ƒò$1pƒò$1½Þ%1t½Þ%17ì%1T7ì%1�€&1à�€&1]R'1$]R'1e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•Éi5 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 E…–ÖÊ  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•Éi5 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 E…–ÖÊ  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•Éi5  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 E…–ÖÊ   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 µÕ~ž,L bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1E…Ê «ëO�Ô  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dEÅ ŠÏO”YÙLÌž—ãcäd¨(¯/mí2²÷w¼< �  F Æ  ‹ Ð P • Õ  Z ^ ž Ÿ ß $d bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ö ÖKKÀÀ)5*54ª4ª>!?!I–I–S T ^�^�hÊhÊrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÖÖKKÀÀ 5 5 ª ª””  ~~óóhhÝÝR R"Ç"Ç$<%<'±'±)&*&,›,›./1…1…3ú3ú5o6o8ä8ä:Y;Y=Î=Î?C@CB¸B¸D-E-G¢G¢IJLŽLŽNOQxQxSíSíUbVbX×X×ZL[L]Á]Á_6`6b«b«d e g•g•i j llnõnõpgqgsÙsÙuKvKx½x½z{}x}xêê�\‚\„΄Άi‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1}}!1Ð}}!1¾!1L¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Oç!1(Oç!1í'"1¤í'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1!Q"1€!Q"1¿‘"1ü¿‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1aÚ"1´ aÚ"1Áþ"10 Áþ"1,#1P,#1o2#1,o2#1ë‹#1˜ë‹#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1�&1Ì�&1íR'1íR'1•e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz×£ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TE…Ó bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz×£ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TE…Ó bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fz×£  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TE…Ó  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ˜¸µÕ@` bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1S€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1“Þ^©)tôà   €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Û ÛTTÍÍ)F*F4¿4¿>8?8I¯I¯S$T$^—^—h i s}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÛÛTTÍÍ F F ¿ ¿88±±**££•• ‡ ‡"#%y%y'ò'ò)k*k,ä,ä.]/]1Ö1Ö3O4O6È6È8A9A;º;º=3>3@¬@¬B%C%EžEžGHJ�J�LMO~O~QõQõSlTlVãVãXZYZ[Ñ[Ñ]H^H`¿`¿b4c4e©e©ghj“j“lmo{o{qîqîsatavÔvÔxGyG{º{º}-~-€ € ‚ƒ…†…†‡ù‡ù‰kŠkŒ݌ݎO�O‘Á‘Á“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1~!1Ð~!1Ÿ¾!1LŸ¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Óç!1(Óç!1q("1¤q("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1¥Q"1€¥Q"1C’"1üC’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1CÚ"1´ CÚ"1£þ"10 £þ"1å+#1På+#1o2#1,o2#1Œ#1˜Œ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1ó€&1Ìó€&1ÃR'1ÃR'1ke'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼Z& b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O�Ø«ë bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼Z& b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O�Ø«ë bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼Z&  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O�Ø«ë  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 µÕˆ¨ @ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1W€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1—äd±1~þç '  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á á[[ÑÑ)G*G4½4½>3?3I§I§ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±++¥¥ ™ ™"#%‰%‰'ÿ'ÿ)u*u,ë,ë.a/a1×1×3M4M6Ã6Ã8999;¯;¯=%>%@›@›BCE‡E‡GýGýIsJsLçLçN[O[QÏQÏSCTCV·V·X+Y+[Ÿ[Ÿ]^`‡`‡bûbûdoeogãgãiWjWlËlËn?o?q³q³s't'v›v›xy{ƒ{ƒ}÷}÷k€k‚߂߄S…S‡LJlj;Š;Œ¯Œ¯Ž#�#‘—‘—“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1}}!1Ð}}!1¾!1L¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Oç!1(Oç!1í'"1¤í'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1!Q"1€!Q"1¿‘"1ü¿‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1=Ú"1´ =Ú"1�þ"10 �þ"1ß+#1Pß+#1÷4#1,÷4#1uŒ#1˜uŒ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1±€&1̱€&1�R'1�R'1)e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÕ¡ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UE…Ò bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÕ¡ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UE…Ò bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1FzÕ¡  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UE…Ò  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x˜µÕ= bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1žÞ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1?¿ ŠÕU  N Ž  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÇÇ)>*>4µ4µ>,?,I¡I¡ST^…^…höhörhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbäbädWeWgÊgÊi=j=l°l°n!o!q’q’stvtvtxåxåzV{V}Ç}Ç8€8‚©‚©„…‡Œ‡Œ‰ý‰ý‹nŒnŽߎß�P‘P“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l+i!1~!1Ð~!1¥¾!1L¥¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Ùç!1(Ùç!1w("1¤w("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1«Q"1€«Q"1I’"1üI’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1ý4#1,ý4#1™Œ#1˜™Œ#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1{€&1Ì{€&1KR'1KR'1ód'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1о^* b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 P�Ú¬ì bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1о^* b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 P�Ú¬ì bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1о^*  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 P�Ú¬ì  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 µÕƒ£6V bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¤ä€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1K˘åe²2t ´  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝUUÉÉ)=*=4±4±>%?%I—I—S T ^{^{híhír_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%���� } }"õ"õ$i%i'Ý'Ý)Q*Q,Å,Å.9/91­1­3!4!6•6•8 9 ;};}=ñ=ñ?e@eBÙBÙDMEMGÁGÁI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnãnãpUqUsÇsÇu9v9x«x«z{}�}�€‚s‚s„å„å†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ƒ}!1Ѓ}!1!¾!1L!¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1Uç!1(Uç!1ó'"1¤ó'"1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1'Q"1€'Q"1Å‘"1üÅ‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1aÚ"1´ aÚ"1Áþ"10 Áþ"1,#1P,#1c2#1,c2#1Á‹#1˜Á‹#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%15�&1Ì5�&1S'1S'1­e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ؤ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 YG‡à  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ؤ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 YG‡à  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ؤ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 YG‡à   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 µÕ—·>^ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1F€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1†ÑQšcã˜ Ø  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚRRÊÊ)B*B4º4º>2?2I¨I¨ST^”^”h i s€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""ššŠŠ z z"ò"ò$j%j'â'â)Z*Z,Ò,Ò.J/J1Â1Â3:4:6²6²8*9*;¢;¢=>@’@’B C E‚E‚GúGúIrJrLèLèN^O^QÔQÔSJTJVÀVÀX6Y6[¬[¬]"^"`˜`˜bce„e„gúgúipjplælæn\o\qÒqÒsHtHv¾v¾x4y4{ª{ª} ~ €–€–‚ ƒ …‚…‚‡ø‡ø‰nŠnŒäŒäŽZ�Z‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1û}!1Ðû}!1™¾!1L™¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Íç!1(Íç!1k("1¤k("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1ŸQ"1€ŸQ"1=’"1ü=’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1i2#1,i2#1'Œ#1˜'Œ#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±O b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ZšŒÌî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±O b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ZšŒÌî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±O  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ZšŒÌî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‰©Rr"B bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1N€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ŽÝ]¬,{ûþ >  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94®4®>!?!I”I”ST^z^zhíhír`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?M@MBÀBÀD3E3G¦G¦IJLŒLŒNÿNÿPrQrSåSåUXVXXËXËZ>[>]±]±_$`$b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†T‡T‰ƉƋ8Œ8ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ƒ}!1Ѓ}!1!¾!1L!¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Uç!1(Uç!1ó'"1¤ó'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1'Q"1€'Q"1Å‘"1üÅ‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1CÚ"1´ CÚ"1£þ"10 £þ"1å+#1På+#1ë4#1,ë4#1KŒ#1˜KŒ#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ö¢ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 XF†Þ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ö¢ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 XF†Þ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ö¢  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 XF†Þ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 µÕw—? bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Ï€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1,¬õu¾>‡! a  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÄÄ):*:4°4°>&?&IšIšST^‚^‚höhörjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$L%L'Â'Â)8*8,®,®.$/$1š1š346†6†8ü8ü:r;r=è=è?^@^BÔBÔDJEJGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqvqvsêsêu^v^xÒxÒzF{F}º}º.€.‚¢‚¢„…‡ЇЉþ‰þ‹rŒr޿޿�Z‘Z“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1~!1Ð~!1Ÿ¾!1LŸ¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Íç!1(Íç!1k("1¤k("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1ŸQ"1€ŸQ"1=’"1ü=’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1ý4#1,ý4#1·Œ#1˜·Œ#1™$1$™$1qò$1èqò$1«Þ%1ì«Þ%1€&1Ì€&1ëQ'1ëQ'1“d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³S b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [›‹Ëð0 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³S b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [›‹Ëð0 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³S  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [›‹Ëð0  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ˆ¨3S@` bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Ý€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1JÊ™èh·7� Í  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚMMÀÀ)3*34¦4¦>?IˆIˆSùSù]j^jhÛhÛrLsL} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠýýppããV V"É"É$<%<'¯'¯)"*",•,•./1{1{3î3î5a6a8Ô8Ô:G;G=º=º?+@+BœBœD E G~G~IïIïK`L`NÑNÑPBQBS³S³U$V$X•X•Z[]w]w_è_èaYbYdÊdÊf;g;i¬i¬klnŽnŽpÿpÿrpspuáuáwRxRzÃzÃ|4}4¥¥�‚„‡„‡†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1‰}!1Љ}!1'¾!1L'¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1[ç!1([ç!1ù'"1¤ù'"1y/"1¬y/"19"1(9"19<"1ô9<"1-Q"1€-Q"1Ë‘"1üË‘"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1c2#1,c2#1Œ#1˜Œ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1ù€&1Ìù€&1ÉR'1ÉR'1qe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ù¥ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  I]�@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ù¥ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  I]�@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ù¥  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  I]�@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 µÕ–¶<\ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÎNœjêð 0  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÇÇ)>*>4µ4µ>,?,I¡I¡ST^‹^‹hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbæbæd[e[gÐgÐiEjElºlºn/o/q¤q¤stvŽvŽxy{x{x}í}íb€b‚ׂׄL…L‡Á‡Á‰6Š6Œ«Œ«Ž � ‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1õ}!1Ðõ}!1“¾!1L“¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Çç!1(Çç!1e("1¤e("1[/"1¬[/"1å8"1(å8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1�Ú"1´ �Ú"1ýþ"10 ýþ"1?,#1P?,#1i2#1,i2#1EŒ#1˜EŒ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦D b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Zš àâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦D b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Zš àâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦D  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Zš àâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Šª'Gq‘ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1E€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1…ÖV§'xø U  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßTTÉÉ)>*>4³4³>)?)I�I�ST^…^…hùhùroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ‰‰þþs s"è"è$]%]'Ò'Ò)G*G,¼,¼.1/11¦1¦346�6�89;z;z=ï=ï?d@dBÙBÙDNENGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒyŒyŽîŽî�c‘c“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1‰}!1Љ}!1'¾!1L'¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1Uç!1(Uç!1ó'"1¤ó'"1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1'Q"1€'Q"1Å‘"1üÅ‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#1ñ4#1,ñ4#1«Œ#1˜«Œ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|×£ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  J\œþ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|×£ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  J\œþ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|×£  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  J\œþ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 µÕ{›&F bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ŽÎ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18¸†ÔT¢"~ ¾  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙOOÄÄ)9*94®4®>#?#I–I–S T ^|^|hïhïrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰ÿÿuuëëa a"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnänäpWqWsÊsÊu=v=x°x°z#{#}–}– € ‚|‚|„ï„ï†b‡b‰Õ‰Õ‹HŒH޻޻�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1û}!1Ðû}!1™¾!1L™¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Çç!1(Çç!1e("1¤e("1[/"1¬[/"1å8"1(å8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#1÷4#1,÷4#1ÕŒ#1˜ÕŒ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨H b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [›¡áä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨H b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [›¡áä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨H  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [›¡áä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �­VvJj bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1–Ö€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1IÉšëk¼<¦ æ  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛPPÅÅ):*:4¯4¯>$?$I—I—S T ^{^{híhírasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹uuêê_ _"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I$J$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b¡b¡deg…g…i÷i÷kilinÛnÛpMqMs¿s¿u1v1x£x£z{}‡}‡ùù�k‚k„݄݆Q‡Q‰ĉċ7Œ7ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1�}!1Ð�}!1-¾!1L-¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1aç!1(aç!1ÿ'"1¤ÿ'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"13Q"1€3Q"1Ñ‘"1üÑ‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1c2#1,c2#1ÕŒ#1˜ÕŒ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1á€&1Ìá€&1±R'1±R'1Ye'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Ú¦ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 NE…Ê  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Ú¦ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 NE…Ê  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Ú¦  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 NE…Ê   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 •µKk:Z bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€Û[´4� ¢ â  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff à àVVÌÌ)B*B4¸4¸>,?,I I ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffààZZÔÔ N N È ÈBB¼¼66°°&&œœ ˆ ˆ"þ"þ$t%t'ê'ê)`*`,Ö,Ö.L/L1Â1Â38486®6®8$9$;š;š=>@„@„BøBøDlElGàGàITJTLÈLÈN3?3IªIªS!T!^“^“hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâI[J[LÒLÒNIOIQÀQÀS7T7V®V®X%Y%[œ[œ]^`Š`ŠbcexexgïgïifjflÝlÝnOoOqÁqÁs3t3v¥v¥xy{‰{‰}û}ûm€m‚߂߄Q…Q‡ÇÉ5Š5Œ§Œ§Ž�‘‹‘‹“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1�}!1Ð�}!1-¾!1L-¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1[ç!1([ç!1ù'"1¤ù'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1-Q"1€-Q"1Ë‘"1üË‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1ý4#1,ý4#1k�#1˜k�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1i€&1Ìi€&19R'19R'1ád'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ؤ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  ME…Ê  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ؤ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  ME…Ê  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ؤ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  ME…Ê   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~ž6V,L bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1™Ù€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1fæ?¿˜ñq;{ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞRRÆÆ):*:4®4®> ? I’I’ST^v^vhèhèrZsZ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žž††úún n"â"â$V%V'Ê'Ê)>*>,²,².&/&1š1š346‚6‚8ö8ö:j;j=Þ=Þ?P@PBÂBÂD4E4G¦G¦IJLŠLŠNüNüPnQnSàSàURVRXÄXÄZ6[6]¨]¨_`bŒbŒdþdþfpgpiâiâkTlTnÆnÆp8q8sªsªuvxŽxŽz{}r}rää�V‚V„ȄȆ:‡:‰«‰«‹ŒŽ�Ž��þ�þ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1ï}!1Ðï}!1�¾!1L�¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Áç!1(Áç!1_("1¤_("1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1ý4#1,ý4#1)�#1˜)�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 E…Ê �Ð bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 E…Ê �Ð bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 E…Ê �Ð  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ž®Llg‡ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1—×€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1\Ü3³ Šáa÷ 7 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ú ÚOOÄÄ)9*94®4®>%?%IšIšST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÚÚOOÄÄ 9 9 ® ®##˜˜  ‚‚÷÷llááV V"Ë"Ë$@%@'µ'µ)***,Ÿ,Ÿ./1‰1‰3þ3þ5s6s8è8è:];]=Ò=Ò?G@GB¼B¼D1E1G¦G¦IJL’L’NOQ|Q|SñSñUfVfXÛXÛZP[P]Å]Å_:`:b¯b¯d$e$g™g™ijlƒlƒnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1•}!1Е}!13¾!1L3¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1aç!1(aç!1ÿ'"1¤ÿ'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"13Q"1€3Q"1Ñ‘"1üÑ‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#1i2#1,i2#1ƒ�#1˜ƒ�#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%15�&1Ì5�&1S'1S'1­e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Û§ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @F†Í  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Û§ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @F†Í  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Û§  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @F†Í   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ’²Pp4T bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ädÈH¬,P� €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff ß ßXXÑÑ)J*J4Ã4Ã>7@°@°B)C)E¢E¢GHJ”J”L M O‚O‚QùQùSpTpVçVçX^Y^[Õ[Õ]L^L`Ã`Ãb8c8e­e­g"h"j—j—l m ooqòqòsetevØvØxKyK{¾{¾}1~1€¤€¤‚ƒ…ЅЇý‡ý‰oŠoŒáŒáŽS�S‘őœ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1é}!1Ðé}!1‡¾!1L‡¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1µç!1(µç!1S("1¤S("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1‡Q"1€‡Q"1%’"1ü%’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1�Ú"1´ �Ú"1ýþ"10 ýþ"1?,#1P?,#1i2#1,i2#1G�#1˜G�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�.ú b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 F†Ë —× bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�.ú b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 F†Ë —× bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�.ú  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 F†Ë —×  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 „¤;g‡ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€âbÄD¦&P €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk å å__ÕÕ)K*K4Á4Á>7?7I«I«ST^“^“his{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkåå__ÙÙ S S Í ÍGGÁÁ;;µµ//©©## � �"#%�%�'(*y*y,ï,ï.e/e1Û1Û3Q4Q6Ç6Ç8=9=;³;³=)>)@Ÿ@ŸBCE‹E‹GHJwJwLëLëN_O_QÓQÓSGTGV»V»X/Y/[£[£]^`‹`‹bÿbÿdsesgçgçi[j[lÏlÏnCoCq·q·s+t+vŸvŸxy{‡{‡}û}ûo€o‚ã‚ã„W…W‡ˇˉ?Š?Œ³Œ³Ž'�'‘›‘›“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1�}!1Ð�}!1-¾!1L-¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1aç!1(aç!1ÿ'"1¤ÿ'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"13Q"1€3Q"1Ñ‘"1üÑ‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#15#1,5#1+Ž#1˜+Ž#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ù¥ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @F†Î bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ù¥ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @F†Î bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ù¥  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @F†Î  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �¡Ii2R bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¤ä€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�tôXØ<¼ô4 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ý ÝTTËË)B*B4¹4¹>0?0I¥I¥ST^‰^‰húhúrlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÝÝTTËË B B ¹ ¹00§§••  ƒƒúúq q"è"è$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n%o%q–q–stvxvxxéxézZ{Z}Ë}Ë<€<‚­‚­„…‡�‡�‰ŠŒrŒrŽãŽã�T‘T“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1é}!1Ðé}!1‡¾!1L‡¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1»ç!1(»ç!1Y("1¤Y("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1�Q"1€�Q"1+’"1ü+’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1�Ú"1´ �Ú"1ýþ"10 ýþ"1?,#1P?,#1 5#1, 5#1é�#1˜é�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’2þ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 E…Ê —× bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’2þ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 E…Ê —× bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’2þ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 E…Ê —×  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‡§>^f† bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¢â€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ˆjêLÌ.®²ò €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áYYÍÍ)A*A4µ4µ>)?)I›I›S T ^^hñhñrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‘‘  � �"ù"ù$m%m'á'á)U*U,É,É.=/=1±1±3%4%6™6™8 9 ;�;�=õ=õ?i@iBÝBÝDQEQGÅGÅI9J9L«L«NOQ�Q�STVsVsXåXåZW[W]É]É_;`;b­b­deg‘g‘ijlulunçnçpYqYsËsËu=v=x¯x¯z!{!}“}“€‚w‚w„é„é†[‡[‰͉͋?Œ?ޱޱ�#‘#“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1›}!1Л}!19¾!1L9¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1gç!1(gç!1("1¤("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"19Q"1€9Q"1ב"1üב"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#1o2#1,o2#17Ž#1˜7Ž#1á$1$á$1¹ò$1è¹ò$1óÞ%1ìóÞ%1G�&1ÌG�&1S'1S'1¿e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kܨ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 CE…Ï bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kܨ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 CE…Ï bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kܨ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 CE…Ï  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 …¥Ii: bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ïoÞ^ÍM@ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÎÎ)F*F4¾4¾>6?6I¬I¬S"T"^˜^˜his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$n%n'æ'æ)^*^,Ö,Ö.N/N1Æ1Æ3>4>6¶6¶8.9.;¦;¦=>@–@–BCE†E†GþGþIvJvLìLìNbObQØQØSNTNVÄVÄX:Y:[°[°]&^&`œ`œbceˆeˆgþgþitjtlêlên`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1Ý}!1ÐÝ}!1{¾!1L{¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1¯ç!1(¯ç!1M("1¤M("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1�Q"1€�Q"1’"1ü’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1=Ú"1´ =Ú"1�þ"10 �þ"1ß+#1Pß+#1o2#1,o2#1û�#1˜û�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…#ï b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I‰ÎšÚ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…#ï b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I‰ÎšÚ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…#ï  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I‰ÎšÚ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 µÕy™ ) bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ímÚZÇGÀ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞSSÈÈ)=*=4²4²>%?%I˜I˜S T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽxxííb b"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?Q@QBÄBÄD7E7GªGªIJL�L�NOQvQvSéSéU\V\XÏXÏZB[B]µ]µ_(`(b›b›deg�g�iôiôkglgnÚnÚpMqMsÀsÀu3v3x¦x¦z{}Œ}Œÿÿ�r‚r„å„å†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1•}!1Е}!13¾!1L3¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1gç!1(gç!1("1¤("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"19Q"1€9Q"1ב"1üב"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#15#1,5#1ëŽ#1˜ëŽ#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÚ¦ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 BE…Ð bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÚ¦ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 BE…Ð bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÚ¦  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 BE…Ð  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 |œJj&F bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¯ï€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¼<«+š‰ ¯ï €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜRRÈÈ)>*>4´4´>*?*IžIžST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒxxîîd d"Ú"Ú$P%P'Æ'Æ)<*<,²,².(/(1ž1ž346Š6Š89;v;v=ì=ì?b@bBØBØDNENGÄGÄI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1ã}!1Ðã}!1�¾!1L�¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1µç!1(µç!1S("1¤S("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1‡Q"1€‡Q"1%’"1ü%’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#15#1,5#1©Ž#1˜©Ž#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1-€&1Ì-€&1ýQ'1ýQ'1¥d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡'ó b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 HˆÍ ›Û bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡'ó b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 HˆÍ ›Û bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡'ó  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 HˆÍ ›Û  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 µÕ€ 0P bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1­í€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1´4¡!Ž{ûm­ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞQQÄÄ)7*74ª4ª>?IŒIŒSýSý]n^nhßhßrPsP} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽttççZ Z"Í"Í$@%@'³'³)&*&,™,™. / 113ò3ò5e6e8Ø8Ø:K;K=¾=¾?/@/B B DEG‚G‚IóIóKdLdNÕNÕPFQFS·S·U(V(X™X™Z [ ]{]{_ì_ìa]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ü†üˆl‰l‹Ü‹Ü�LŽL�¼�¼’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Å}!1�Å}!1c¾!1c¾!1³Â!1$³Â!1‘ç!1�‘ç!1/("1/("1…,"1$…,"1cQ"1�cQ"1’"1’"1W–"1$W–"1sÚ"1�sÚ"1Óþ"1üÓþ"1/0#1 /0#1û~#1| û~#1í$1Ü í$1Åò$1  Åò$1ÿÞ%1¤ ÿÞ%1#�&1„ #�&1óR'1ÈóR'1›e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1%Y]‘îºÏ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1%Y]‘îºÏ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1%Y]‘îºÏ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations13S  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww î îeeÜÜ)S*S4Ê4Ê>???I´I´S)T)^ž^žhisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwîîeeÜÜ S S Ê ÊAA¸¸//¦¦””  ‚ ‚"ù"ù$p%p'ç'ç)^*^,Õ,Õ.L/L1Ã1Ã3:4:6±6±8(9(;Ÿ;Ÿ=>@�@�BCE{E{GòGòIgJgLÜLÜNQOQQÆQÆS;T;V°V°X%Y%[š[š]^`„`„bùbùdnengãgãiXjXlÍlÍnBoBq·q·s,t,v¡v¡xy{‹{‹}~€u€u‚ê‚ê„_…_‡Ô‡Ô‰IŠIŒ¾Œ¾Ž3�3‘¨‘¨“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ý}!1�Ý}!1{¾!1 {¾!1_Å!1_Å!1©ç!1€©ç!1G("1üG("11/"11/"1{Q"1p{Q"1’"1ì’"1™"1ô™"1ùØ"1`ùØ"1Yý"1ÌYý"1/0#1ì/0#1û~#1L û~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1ç€&1T ç€&1·R'1˜·R'1_e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Eyã b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1©é€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Eyã b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1©é€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Eyã b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1©é €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1µÕ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ì ìaaÖÖ)K*K4À4À>4?4I¨I¨ST^�^�hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttììddÜÜ T T Ì ÌDD¼¼44¬¬!!––  € €"õ"õ$j%j'ß'ß)T*T,É,É.>/>1³1³3(4(6�6�89;‡;‡=ü=ü?q@qBæBæD[E[GÐGÐIDJDL¸L¸N,O,Q Q STVˆVˆXüXüZp[p]ä]ä_X`XbÌbÌd@e@g´g´i(j(lœlœnoq„q„søsøulvlxàxàzT{T}È}È<€<‚°‚°„$…$‡˜‡˜‰ Š Œ‚Œ‚Ž÷Ž÷�l‘l“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Å}!1�Å}!1c¾!1c¾!1³Â!1$³Â!1—ç!1�—ç!15("15("1…,"1$…,"1iQ"1�iQ"1’"1’"1W–"1$W–"1Ú"1�Ú"1ßþ"1üßþ"1/0#1 /0#1û~#1| û~#1Õ$1Ü Õ$1­ò$1  ­ò$1çÞ%1¤ çÞ%1·€&1„ ·€&1‡R'1ȇR'1/e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1&Z[�ê¶Ò b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1&Z[�ê¶Ò b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1&Z[�ê¶Ò b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1:Z  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ì ìbb××)L*L4Á4Á>4?4I§I§ST^�^�hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvììbbØØ N N Ä Ä::°°&&œœˆˆþþt t"ê"ê$_%_'Ô'Ô)I*I,¾,¾.3/31¨1¨346’6’89;|;|=ñ=ñ?f@fBÛBÛDPEPGÅGÅI8J8L«L«NOQ‘Q‘STVwVwXêXêZ][]]Ð]Ð_C`Cb¶b¶d)e)gœgœijl‚l‚nõnõphqhsÛsÛuNvNxÁxÁz4{4}§}§€‚�‚�„…‡s‡s‰æ‰æ‹YŒYŽÌŽÌ�?‘?“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ý}!1�Ý}!1{¾!1 {¾!1YÅ!1YÅ!1¯ç!1€¯ç!1M("1üM("1+/"1+/"1�Q"1p�Q"1’"1ì’"1ý˜"1ôý˜"1ùØ"1`ùØ"1Yý"1ÌYý"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Eyå b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¨è€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Eyå b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¨è€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Eyå b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¨è €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1µÕ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ê ê__ÔÔ)I*I4¾4¾>1?1I¤I¤ST^ˆ^ˆhúhúrlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttêê``ÖÖ L L  Â88®®$$šš„„ùùn n"ã"ã$X%X'Í'Í)B*B,·,·.,/,1¡1¡346‹6‹89;u;u=ê=ê?_@_BÔBÔDIEIG¾G¾I1J1L¤L¤NOQŠQŠSýSýUpVpXãXãZV[V]É]É_<`v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰ωϋBŒB޵޵�(‘(“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1­}!1�­}!1K¾!1 K¾!1³Â!1³Â!1yç!1€yç!1("1ü("1…,"1…,"1KQ"1pKQ"1é‘"1ìé‘"1W–"1ôW–"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1ù€&1T ù€&1ÉR'1˜ÉR'1qe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽë· b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1C€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽë· b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1C€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽë· b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1C €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1µÕ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww ñ ñggÝÝ)S*S4Ç4Ç>;?;I¯I¯S#T#^—^—h i s}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwññkkåå _ _ Ù ÙSSÍÍGGÁÁ77­­## ™ ™"#%…%…'û'û)q*q,ç,ç.]/]1Ó1Ó3I4I6¿6¿8595;«;«=>@“@“BCE{E{GïGïIcJcL×L×NKOKQ¿Q¿S3T3V§V§XY[�[�]^`w`wbëbëd_e_gÓgÓiGjGl»l»n/o/q£q£stv‹v‹xÿxÿzs{s}ç}ç[€[‚ςτC…C‡µ‡µ‰(Š(Œ›Œ›Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1ã}!1�ã}!1�¾!1 �¾!1GÅ!1GÅ!1¯ç!1€¯ç!1M("1üM("1/"1/"1�Q"1p�Q"1’"1ì’"1ë˜"1ôë˜"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1 �&1T  �&1ÛR'1˜ÛR'1ƒe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ ì b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1šÚ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ ì b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1šÚ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ ì b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1šÚ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1µÕ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp ç ç^^ÕÕ)L*L4Ã4Ã>:?:I±I±S(T(^š^šh i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppçç^^ÕÕ L L à Ã::±±((ŸŸ�� { {"ò"ò$i%i'à'à)W*W,Î,Î.E/E1¼1¼33436ª6ª8!9!;˜;˜=>@†@†BýBýDtEtGëGëIbJbLÙLÙNPOPQÇQÇS>T>VµVµX,Y,[£[£]^`‘`‘bceegögöimjmlälänVoVqÈqÈs:t:v¬v¬xy{�{�}~€t€t‚æ‚æ„X…X‡ʇʉ<Š<Œ®Œ®Ž � ‘’‘’“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1§}!1�§}!1E¾!1 E¾!1³Â!1³Â!1yç!1€yç!1("1ü("1…,"1…,"1KQ"1pKQ"1é‘"1ìé‘"1W–"1ôW–"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒç³ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1B€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒç³ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1B€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒç³ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1B €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1µÕ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu í íaaÕÕ)I*I4»4»>-?-IŸIŸST^ƒ^ƒhõhõrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuííeeÝÝ U U Í ÍEE½½55­­!!••  } }"ñ"ñ$e%e'Ù'Ù)M*M,Á,Á.5/51©1©346‘6‘89;y;y=ë=ë?]@]BÏBÏDAEAG³G³I%J%L—L—N O Q{Q{SíSíU_V_XÑXÑZC[C]µ]µ_'`'b™b™d e g}g}iïiïkalanÓnÓpEqEs·s·u)v)x›x›z { }}ññ�c‚c„Õ„Õ†E‡E‰¶‰¶‹'Œ'Ž˜Ž˜� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1ã}!1�ã}!1�¾!1 �¾!1MÅ!1MÅ!1µç!1€µç!1S("1üS("1/"1/"1‡Q"1p‡Q"1%’"1ì%’"1ñ˜"1ôñ˜"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1±€&1T ±€&1�R'1˜�R'1)e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚"î b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1›Û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚"î b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1›Û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚"î b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1›Û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1µÕ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp å åZZÏÏ)D*D4¹4¹>.?.I£I£ST^Š^Šhühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppååZZÏÏ D D ¹ ¹..££��wwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇDG?GI¾I¾S3T3^¦^¦hisŠsŠ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssììeeÞÞ W W Ð ÐIIÂÂ;;´´--¦¦ ˜ ˜"#%Š%Š'(*|*|,õ,õ.n/n1ç1ç3`4`6Ù6Ù8R9R;Ë;Ë=D>D@½@½B6C6E¯E¯G(H(JŸJŸLMO�O�QRT{T{VòVòXiYi[à[à]W^W`Î`ÎbCcCe¸e¸g-h-j¢j¢lmoŠoŠqýqýsptpvãvãxVyV{É{É}<~<€¯€¯‚"ƒ"…•…•‡ˆŠxŠxŒêŒêŽ\�\‘ΑΓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1ã}!1�ã}!1�¾!1 �¾!1YÅ!1YÅ!1µç!1€µç!1S("1üS("1+/"1+/"1‡Q"1p‡Q"1%’"1ì%’"1ý˜"1ôý˜"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹)õ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1«ë€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹)õ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1«ë€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹)õ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1«ë €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1µÕ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp ê êddÚÚ)P*P4Æ4Æ>:?:I®I®S"T"^–^–h i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppêêddÞÞ X X Ò ÒLLÆÆ@@ºº44®®(( ¢ ¢"#%’%’'(*~*~,ô,ô.j/j1à1à3V4V6Ì6Ì8B9B;¸;¸=.>.@¤@¤BCE�E�GHJzJzLîLîNbObQÖQÖSJTJV¾V¾X2Y2[¦[¦]^`Ž`Žbcevevgêgêi^j^lÒlÒnFoFqºqºs.t.v¢v¢xy{Š{Š}þ}þr€r‚æ‚æ„Z…Z‡·ΉBŠBŒ¶Œ¶Ž*�*‘ž‘ž“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1¡}!1�¡}!1?¾!1 ?¾!1³Â!1³Â!1sç!1€sç!1("1ü("1…,"1…,"1EQ"1pEQ"1ã‘"1ìã‘"1W–"1ôW–"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Ï€&1T Ï€&1ŸR'1˜ŸR'1Ge'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ä° b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ä° b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ä° b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1µÕ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq è è__ÖÖ)M*M4Ä4Ä>9?9I®I®S!T!^’^’hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqèè__ÖÖ M M Ä Ä;;²²))  ŽŽ | |"ó"ó$j%j'á'á)X*X,Ï,Ï.F/F1½1½34446«6«8"9";™;™=>@‡@‡BþBþDuEuGìGìIaJaLÖLÖNKOKQÀQÀS5T5VªVªXY[”[”] ^ `~`~bñbñddedg×g×iJjJl½l½n.o.qŸqŸstv�v�xòxòzc{c}Ô}ÔE€E‚¶‚¶„'…'‡—‡—‰ŠŒyŒyŽêŽê�[‘[“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1é}!1�é}!1‡¾!1 ‡¾!1SÅ!1SÅ!1µç!1€µç!1S("1üS("1%/"1%/"1‡Q"1p‡Q"1%’"1ì%’"1÷˜"1ô÷˜"1íØ"1`íØ"1Mý"1ÌMý"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹+÷ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ªê€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹+÷ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ªê€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹+÷ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ªê €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1µÕ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp è è``ÔÔ)H*H4¼4¼>.?.I I ST^„^„höhörhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppèè``ØØ P P È È@@¸¸00¨¨  ˜˜ ˆ ˆ"#%t%t'è'è)\*\,Ð,Ð.D/D1¸1¸3,4,6 6 89;ˆ;ˆ=ü=ü?p@pBäBäDXEXGÌGÌI>J>L°L°N"O"Q”Q”STVxVxXêXêZ\[\]Î]Î_@`@b²b²d$e$g–g–ijlzlznìnìp^q^sÐsÐuBvBx´x´z&{&}˜}˜ € ‚|‚|„î„î†`‡`‰Ò‰Ò‹DŒD޶޶�(‘(“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1›}!1�›}!19¾!1 9¾!1¿Â!1¿Â!1mç!1€mç!1 ("1ü ("1‘,"1‘,"1?Q"1p?Q"1Ý‘"1ìÝ‘"1c–"1ôc–"1íØ"1`íØ"1Mý"1ÌMý"1#0#1ì#0#1ï~#1L ï~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1G�&1T G�&1S'1˜S'1¿e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆå± b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Z€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆå± b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Z€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆå± b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Z €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1µÕ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo ç ç__××)O*O4Ç4Ç>=?=I³I³S)T)^Ÿ^Ÿhis‹s‹} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooçç__×× O O Ç Ç??··//§§—— ‡ ‡"ÿ"ÿ$w%w'ï'ï)g*g,ß,ß.W/W1Ï1Ï3G4G6¿6¿8797;¯;¯='>'@Ÿ@ŸBCE�E�GHJ}J}LóLóNiOiQßQßSUTUVËVËXAYA[·[·]-^-`£`£bce�e�ghj{j{lñlñngogqÝqÝsStSvÉvÉx?y?{µ{µ}+~+€¡€¡‚ƒ…�…�‡ˆŠyŠyŒïŒïŽe�e‘Û‘Û“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1é}!1�é}!1‡¾!1 ‡¾!15Å!15Å!1»ç!1€»ç!1Y("1üY("1/"1/"1�Q"1p�Q"1+’"1ì+’"1Ù˜"1ôÙ˜"1Ú"1`Ú"1ßþ"1Ìßþ"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”2þ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Í€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”2þ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Í€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”2þ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Í €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ii  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii ß ßTTÉÉ)>*>4±4±>$?$I—I—S T ^}^}hðhðrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiißßUUËË A A · ·--££��yyîîc c"Ø"Ø$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=Ý=Ý?P@PBÃBÃD6E6G©G©IJL�L�NOQuQuSèSèU[V[XÎXÎZA[A]´]´_'`'bšbšd e g€g€ióiókflfnÙnÙpLqLs¿s¿u2v2x¥x¥z{}‹}‹þþ�q‚q„ä„ä†U‡U‰ljNj9Œ9ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1›}!1�›}!19¾!1 9¾!1ÅÂ!1ÅÂ!1gç!1€gç!1("1ü("1—,"1—,"19Q"1p9Q"1ב"1ìב"1i–"1ôi–"1íØ"1`íØ"1Mý"1ÌMý"1#0#1ì#0#1ï~#1L ï~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Ï€&1T Ï€&1ŸR'1˜ŸR'1Ge'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†á­ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1[€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†á­ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1[€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†á­ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1[ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1µÕ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ã ãYYÏÏ)E*E4»4»>/?/I£I£ST^‹^‹hÿhÿrsss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““  õõk k"á"á$W%W'Í'Í)C*C,¹,¹.///1¥1¥346‘6‘89;};}=ó=ó?i@iBßBßDUEUGËGËI?J?L³L³N'O'Q›Q›STVƒVƒX÷X÷Zk[k]ß]ß_S`SbÇbÇd;e;g¯g¯i#j#l—l—n o qqsósóugvgxÛxÛzO{O}Ã}Ã7€7‚«‚«„…‡“‡“‰ŠŒ{Œ{ŽïŽï�c‘c“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1ï}!1�ï}!1�¾!1 �¾!15Å!15Å!1»ç!1€»ç!1Y("1üY("1/"1/"1�Q"1p�Q"1+’"1ì+’"1Ù˜"1ôÙ˜"1‘Ú"1`‘Ú"1ñþ"1Ìñþ"1#0#1ì#0#1ï~#1L ï~#1¥$1¬ ¥$1}ò$1p }ò$1·Þ%1t ·Þ%1'€&1T '€&1÷Q'1˜÷Q'1Ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”4 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ŽÎ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”4 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ŽÎ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”4 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ŽÎ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Xx  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk á áTTÇÇ):*:4«4«>?I�I�SþSþ]o^ohàhàrOsO} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkááWWÍÍ C C ¹ ¹//¥¥‘‘wwêê] ]"Ð"Ð$C%C'¶'¶))*),œ,œ./1‚1‚3õ3õ5h6h8Û8Û:N;N=¿=¿?0@0B¡B¡DEGƒGƒIôIôKeLeNÖNÖPGQGS¸S¸U)V)XšXšZ [ ]|]|_í_ía^b^dÏdÏf@g@i±i±k"l"n“n“pqsusuuæuæwWxWzÈzÈ|9}9ªª�‚„Œ„Œ†û†ûˆk‰k‹Û‹Û�KŽK�»�»’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1•}!1�•}!13¾!1 3¾!1³Â!1³Â!1gç!1€gç!1("1ü("1…,"1…,"19Q"1p9Q"1ב"1ìב"1W–"1ôW–"1çØ"1`çØ"1Gý"1ÌGý"10#1ì0#1é~#1L é~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…â® b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…â® b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…â® b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1µÕ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk â âYYÐÐ)G*G4¾4¾>3?3I¨I¨ST^’^’his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkââYYÐÐ G G ¾ ¾55¬¬##ššˆˆÿÿv v"í"í$d%d'Û'Û)R*R,É,É.@/@1·1·3.4.6¥6¥89;“;“= > @�@�BøBøDoEoGæGæI[J[LÐLÐNEOEQºQºS/T/V¤V¤XY[Ž[Ž]^`x`xbíbídbebg×g×iLjLlÁlÁn6o6q«q«s t v•v•x y {{}ô}ôi€i‚Þ‚Þ„S…S‡ȇȉ=Š=Œ²Œ²Ž'�'‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ï}!1�ï}!1�¾!1 �¾!1MÅ!1MÅ!1Áç!1€Áç!1_("1ü_("1/"1/"1“Q"1p“Q"11’"1ì1’"1ñ˜"1ôñ˜"1Ú"1`Ú"1ßþ"1Ìßþ"10#1ì0#1é~#1L é~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�; b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¨è€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�; b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¨è€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�; b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¨è €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Nn  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àUUÊÊ)?*?4´4´>(?(IœIœST^„^„høhørlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  ŠŠÿÿt t"é"é$^%^'Ó'Ó)H*H,½,½.2/21§1§346‘6‘89;{;{=ð=ð?e@eBÚBÚDOEOGÄGÄI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡Œ‡Œ‰ŠŒvŒvŽëŽë�`‘`“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1•}!1�•}!13¾!1 3¾!1³Â!1³Â!1aç!1€aç!1ÿ'"1üÿ'"1…,"1…,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1W–"1ôW–"1çØ"1`çØ"1Gý"1ÌGý"10#1ì0#1é~#1L é~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OƒÞª b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OƒÞª b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OƒÞª b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1µÕ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj à àVVËË)@*@4µ4µ>(?(I›I›ST^�^�hôhôrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjààVVÌÌ B B ¸ ¸..¤¤��||òòh h"Þ"Þ$S%S'È'È)=*=,²,².'/'1œ1œ346†6†8û8û:p;p=å=å?Z@ZBÏBÏDDEDG¹G¹I,J,LŸLŸNOQ…Q…SøSøUkVkXÞXÞZQ[Q]Ä]Ä_7`7bªbªdeg�g�ijlvlvnénép\q\sÏsÏuBvBxµxµz({(}›}›€‚�‚�„ô„ô†g‡g‰Ú‰Ú‹MŒMŽÀŽÀ�3‘3“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1ï}!1�ï}!1�¾!1 �¾!1MÅ!1MÅ!1Áç!1€Áç!1_("1ü_("1/"1/"1“Q"1p“Q"11’"1ì1’"1ñ˜"1ôñ˜"1—Ú"1`—Ú"1÷þ"1Ì÷þ"10#1ì0#1é~#1L é~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1§ç€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1§ç€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1§ç €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1e…  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞSSÈÈ)=*=4²4²>%?%I˜I˜S T ^|^|hîhîr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽxxííb b"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†P‡P‰ÉË6Œ6ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1�}!1��}!1-¾!1 -¾!1¹Â!1¹Â!1aç!1€aç!1ÿ'"1üÿ'"1‹,"1‹,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1]–"1ô]–"1OÚ"1`OÚ"1¯þ"1̯þ"10#1ì0#1é~#1L é~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ß« b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1W€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ß« b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1W€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ß« b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1W €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1 @  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk å å[[ÑÑ)G*G4»4»>/?/I£I£ST^‹^‹hÿhÿrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkåå__ÙÙ S S Í ÍGGÁÁ;;µµ++¡¡ � �"#%y%y'ï'ï)e*e,Û,Û.Q/Q1Ç1Ç3=4=6³6³8)9);Ÿ;Ÿ=>@‡@‡BûBûDoEoGãGãIWJWLËLËN?O?Q³Q³S'T'V›V›XY[ƒ[ƒ]÷]÷_k`kbßbßdSeSgÇgÇi;j;l¯l¯n#o#q—q—s t vvxóxózg{g}Û}ÛO€O‚ÂÄ7…7‡©‡©‰ŠŒ�Œ�Ž�‘u‘u“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1õ}!1�õ}!1“¾!1 “¾!1)Å!1)Å!1Çç!1€Çç!1e("1üe("1û."1û."1™Q"1p™Q"17’"1ì7’"1͘"1ô͘"1…Ú"1`…Ú"1åþ"1Ìåþ"10#1ì0#1é~#1L é~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1í€&1T í€&1½R'1˜½R'1ee'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦D b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‹Ë€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦D b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‹Ë€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦D b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‹Ë €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Uu  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I¥I¥ST^Ž^Žhisrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLÍLÍNDODQ»Q»S2T2V©V©X Y [—[—]^`…`…bübüdsesgêgêiajalØlØnJoJq¼q¼s.t.v v xy{„{„}ö}öh€h‚Ú‚Ú„L…L‡¾‡¾‰0Š0Œ¢Œ¢Ž�‘†‘†“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1‰}!1�‰}!1'¾!1 '¾!1¹Â!1¹Â!1[ç!1€[ç!1ù'"1üù'"1‹,"1‹,"1-Q"1p-Q"1Ë‘"1ìË‘"1]–"1ô]–"1aÚ"1`aÚ"1Áþ"1ÌÁþ"10#1ì0#1é~#1L é~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1c€&1T c€&13R'1˜3R'1Ûd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Û§ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1X€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Û§ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1X€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Û§ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1X €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations15U  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áUUÉÉ)=*=4¯4¯>!?!I“I“ST^w^whéhérYsY} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‰‰ýýq q"å"å$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b�b�dÿdÿfqgqiãiãkUlUnÇnÇp9q9s«s«uvx�x�z{}s}såå�W‚W„ɄɆ9‡9‰ª‰ª‹ŒŽŒŽŒ�ý�ý’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1õ}!1�õ}!1“¾!1 “¾!1)Å!1)Å!1Çç!1€Çç!1e("1üe("1û."1û."1™Q"1p™Q"17’"1ì7’"1͘"1ô͘"1�Ú"1`�Ú"1ýþ"1Ìýþ"10#1ì0#1é~#1L é~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦F b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‹Ë€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦F b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‹Ë€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦F b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‹Ë €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1p�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ù ÙNNÃÃ)8*84­4­>"?"I—I—S T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÙÙNNÃà 8 8 ­ ­""——  ��öökkààU U"Ê"Ê$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1‰}!1�‰}!1'¾!1 '¾!1§Â!1§Â!1[ç!1€[ç!1ù'"1üù'"1y,"1y,"1-Q"1p-Q"1Ë‘"1ìË‘"1K–"1ôK–"1OÚ"1`OÚ"1¯þ"1̯þ"10#1ì0#1ã~#1L ã~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1)�&1T )�&1ùR'1˜ùR'1¡e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kܨ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kܨ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kܨ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1%E  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg à àYYÒÒ)K*K4Ä4Ä>;?;I²I²S'T'^œ^œhis€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggààYYÒÒ K K Ä Ä==¶¶//¨¨!!šš Œ Œ"#%~%~'÷'÷)p*p,é,é.b/b1Û1Û3T4T6Í6Í8F9F;¿;¿=8>8@±@±B*C*E£E£GHJ“J“L M O�O�QøQøSoToVæVæX]Y][Ô[Ô]K^K`Â`Âb7c7e¬e¬g!h!j–j–l m o€o€qóqósftfvÙvÙxLyL{¿{¿}2~2€¥€¥‚ƒ…‹…‹‡ü‡ü‰nŠnŒàŒàŽR�R‘đē b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1û}!1�û}!1™¾!1 ™¾!1AÅ!1AÅ!1Çç!1€Çç!1e("1üe("1/"1/"1™Q"1p™Q"17’"1ì7’"1å˜"1ôå˜"1aÚ"1`aÚ"1Áþ"1ÌÁþ"10#1ì0#1ã~#1L ã~#1É$1¬ É$1ñð$1p ñð$1ÏÀ%1t ÏÀ%1ÿ€&1T ÿ€&1õP'1˜õP'1�c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯M b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¥å€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯M b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¥å€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯M b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¥å €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1XT  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Þ ÞXXÎÎ)D*D4º4º>.?.I¢I¢ST^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÞÞXXÒÒ L L Æ Æ@@ºº44®®((¢¢ – –"#%†%†'ü'ü)r*r,è,è.^/^1Ô1Ô3J4J6À6À8696;¬;¬=">"@˜@˜BCE„E„GúGúInJnLâLâNVOVQÊQÊS>T>V²V²X&Y&[š[š]^`‚`‚böbödjejgÞgÞiRjRlÆlÆnDoDq¸q¸s,t,v v xy{ˆ{ˆ}ü}üp€p‚ä‚ä„X…X‡̇̉@Š@Œ´Œ´Ž(�(‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1ƒ}!1�ƒ}!1!¾!1 !¾!1§Â!1§Â!1Uç!1€Uç!1ó'"1üó'"1y,"1y,"1'Q"1p'Q"1Å‘"1ìÅ‘"1K–"1ôK–"1mÚ"1`mÚ"1Íþ"1ÌÍþ"10#1ì0#1ã~#1L ã~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1·€&1T ·€&1‡R'1˜‡R'1/e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ؤ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ؤ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ؤ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Bb  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>-?-I¢I¢ST^Š^Šhûhûrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbåbådXeXgËgËi>j>l±l±n&o&q—q—stvyvyxêxêz[{[}Ì}Ì=€=‚®‚®„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1û}!1�û}!1™¾!1 ™¾!1AÅ!1AÅ!1Íç!1€Íç!1k("1ük("1/"1/"1ŸQ"1pŸQ"1=’"1ì=’"1å˜"1ôå˜"1…Ú"1`…Ú"1åþ"1Ìåþ"10#1ì0#1ã~#1L ã~#1±$1¬ ±$1Ùð$1p Ùð$1·À%1t ·À%1‡€&1T ‡€&1}P'1˜}P'1%c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯O b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¤ä€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯O b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¤ä€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯O b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¤ä €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1}y  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜTTÈÈ)<*<4°4°>"?"I”I”ST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœŒŒ | |"ô"ô$h%h'Ü'Ü)P*P,Ä,Ä.8/81¬1¬3 4 6”6”89;|;|=ð=ð?d@dBØBØDLELGÀGÀI2J2L¤L¤NOQˆQˆSúSúUlVlXÞXÞZP[P]Â]Â_4`4b¦b¦degŠgŠiüiüknlnnènèpZqZsÌsÌu>v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1ƒ}!1�ƒ}!1!¾!1 !¾!1­Â!1­Â!1Oç!1€Oç!1í'"1üí'"1,"1,"1!Q"1p!Q"1¿‘"1ì¿‘"1Q–"1ôQ–"1·Ø"1`·Ø"1ý"1Ìý"10#1ì0#1Ý~#1L Ý~#1Õ$1¬ Õ$1ýð$1p ýð$1ÛÀ%1t ÛÀ%1;�&1T ;�&11Q'1˜1Q'1Ùc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ù¥ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1T€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ù¥ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1T€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ù¥ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1T €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1µ±  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSÍÍ)E*E4½4½>3?3I©I©ST^¡^¡his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=>@•@•B C E…E…GýGýIsJsLéLéN_O_QÕQÕSKTKVÁVÁX7Y7[­[­]#^#`™`™bce…e…gûgûiqjqlçlçnioiqßqßsUtUvËvËxAyA{·{·}-~-€£€£‚ƒ…�…�‡ˆŠ{Š{ŒñŒñŽg�g‘ݑݓ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1~!1�~!1Ÿ¾!1 Ÿ¾!1#Å!1#Å!1Íç!1€Íç!1k("1ük("1õ."1õ."1ŸQ"1pŸQ"1=’"1ì=’"1ǘ"1ôǘ"1%Ú"1`%Ú"1…þ"1Ô…þ"10#1ô0#1Ý~#1T Ý~#1±$1´ ±$1Ùð$1x Ùð$1·À%1| ·À%1‡€&1\ ‡€&1}P'1 }P'1%c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸V" b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‹Ë€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸V" b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‹Ë€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸V" b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‹Ë €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1¡éj  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10]] Ó ÓRRÇÇ)<*<4¯4¯>"?"I•I•ST^‡^‡húhúrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð]]ÓÓII¿¿ 5 5 « «!!——  ƒƒwwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Û=Û?N@NBÁBÁD4E4G§G§IJL�L�NOQsQsSæSæUYVYXÌXÌZ?[?]²]²_%`%b¤b¤degŠgŠiýiýkplpnãnãpVqVsÉsÉu'?'I›I›ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqƒqƒs÷s÷ukvkxßxßzS{S}Ç}Ç;€;‚¯‚¯„#…#‡—‡—‰ Š ŒŒŽóŽó�g‘g“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L+i!1~!1�~!1Ÿ¾!1 Ÿ¾!1#Å!1#Å!1Óç!1€Óç!1q("1üq("1õ."1õ."1¥Q"1p¥Q"1C’"1ìC’"1ǘ"1ôǘ"1Ú"1`Ú"1sþ"1Ôsþ"10#1ô0#1Ý~#1T Ý~#1Ÿ$1´ Ÿ$1Çð$1x Çð$1¥À%1| ¥À%1!€&1\ !€&1P'1 P'1¿b'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸X$ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‹Ë€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸X$ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‹Ë€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸X$ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‹Ë €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1¡éX   bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Õ ÕPPÃÃ)6*64§4§>?I‰I‰ST^w^whèhèrWsW} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……ssææY Y"Ì"Ì$?%?'²'²)%*%,˜,˜. / 1~1~3ñ3ñ5d6d8×8×:J;J=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éafbfd×d×fHgHi¹i¹k*l*n›n›p q s}s}uîuîw_x_zÐzÐ|A}A²²�#‚#„”„”†‡‰s‰s‹ã‹ã�SŽS�Ã�Ã’ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„%i!1M}!1M}!1ë½!1|ë½!1OÃ!1„OÃ!1…Æ!1ð…Æ!1Ð!1|Ð!1EÓ!1HEÓ!1ç!1Ôç!1·'"1P·'"1!-"1X!-"1W0"1ÄW0"1á9"1Pá9"1="1="1ëP"1¨ëP"1‰‘"1$ ‰‘"1ó–"1, ó–"1)š"1˜ )š"1³£"1$ ³£"1é¦"1ð é¦"1+Ô"1| +Ô"1Ú"1XÚ"1ßþ"1Ôßþ"1!,#1ô!,#11:#1Ð1:#1ín#1\ín#1q�#1€q�#1o˜#1¬o˜#1É$18É$1¡ò$1ü¡ò$1ÛÞ%1ÛÞ%1Uì%1àUì%1ÿ€&1lÿ€&1ÏR'1°ÏR'1we'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¯{ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1·÷ bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¯{ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1·÷  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¯{  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1·÷ bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �­]}Jj bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1G‡ÎU•?Ü  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dGÇŽÕUœcã*ªñq¸8�™HÈhè�7·ÖV  † �  d ä + « ò r ¹ ù @ € Ç NŽ/o bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa ± ±UUÕÕ)K*K4Á4Á>7?7I«I«ST^Ÿ^Ÿhis‡s‡} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÛÛUUÏÏ I I ™ ™����%%ŸŸ “ “" # %�%�'(*y*y,ï,ï.e/e1Û1Û3Q4Q6Ç6Ç8=9=;³;³=)>)@Ÿ@ŸBCE‹E‹GHJwJwLëLëN_O_QÓQÓSGTGV»V»X/Y/[£[£]^`‹`‹bÿbÿdsesgçgçi[j[lÏlÏnOoOqÃqÃs7t7v«v«xy{“{“}~€{€{‚ï‚ï„c…c‡ׇ׉KŠKŒ¿Œ¿Ž3�3‘§‘§“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1±¾!1d±¾!1eÅ!1leÅ!1ïÎ!1èïÎ!1%Ò!1´%Ò!1åç!1@åç!1ƒ("1¼ƒ("17/"1Ä7/"1Á8"1@Á8"1÷;"1 ÷;"1·Q"1˜·Q"1U’"1U’"1 ™"1  ™"1“¢"1˜ “¢"1É¥"1d É¥"1 Ó"1ð  Ó"1+Ú"1Ì +Ú"1‹þ"1H ‹þ"1Í+#1hÍ+#1U:#1DU:#1o#1Ðo#1 Ž#1ô Ž#1 ™#1  ™#1Ï$1¬Ï$1§ò$1p§ò$1áÞ%1táÞ%1[ì%1T[ì%1)�&1à)�&1ùR'1$ùR'1¡e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ôt@ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G‡‘ÑÎ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ôt@ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G‡‘ÑÎ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ôt@  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G‡‘ÑÎ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ·×r’2 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1NŽÜEl¬ü<  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dNΜêj¸8†TÔ"¢ðpÜ\À@•NÎ�‡`à0 °  € Ð P   p ð – Ö > ~ Ì \œì, bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ý ÝVVÑÑ) * 4Ã4Ã>7@°@°B)C)E¢E¢GHJ”J”L M O‚O‚QùQùSpTpVçVçX^Y^[Õ[Õ]L^L`Ã`Ãb8c8e­e­g"h"j—j—l m o‹o‹qþqþsqtqväväxWyW{Ê{Ê}=~=€°€°‚#ƒ#…–…–‡ ˆ Š{Š{ŒíŒíŽ_�_‘ёѓ  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„%i!1G}!1G}!1å½!1|å½!1OÃ!1„OÃ!1…Æ!1ð…Æ!1Ð!1|Ð!1EÓ!1HEÓ!1ç!1Ôç!1·'"1P·'"1!-"1X!-"1W0"1ÄW0"1á9"1Pá9"1="1="1ëP"1¨ëP"1‰‘"1$ ‰‘"1ó–"1, ó–"1)š"1˜ )š"1³£"1$ ³£"1é¦"1ð é¦"1+Ô"1| +Ô"1OÚ"1XOÚ"1¯þ"1Ô¯þ"1ñ+#1ôñ+#1I:#1ÐI:#1o#1\o#1é�#1€é�#1ç˜#1¬ç˜#1±$18±$1‰ò$1ü‰ò$1ÃÞ%1ÃÞ%1=ì%1à=ì%1‡€&1l‡€&1WR'1°WR'1ÿd'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¬x b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1·÷ bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¬x b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1·÷  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¬x  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1·÷ bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 “³,L6V bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1NŽÜh¨ô4 Î  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dNΜêj¸8†TÔ"¢ðp¼<ˆTÔ   ì l Õ U ¸ 8 ¤ $ „  s ó P Ð B   ^ ¬ ì 8xÄ¿ÿ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa ® ®QQÍÍ)A*A4µ4µ>)?)I›I›S T ^‹^‹hýhýroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A Ž Ž~~öönn‰‰ y y"ñ"ñ$m%m'á'á)U*U,É,É.=/=1±1±3%4%6™6™8 9 ;�;�=õ=õ?i@iBÝBÝDQEQGÅGÅI9J9L«L«NOQ�Q�STVsVsXåXåZW[W]É]É_;`;b­b­deg‘g‘ijlulunónópeqes×s×uIvIx»x»z-{-}Ÿ}Ÿ€‚ƒ‚ƒ„õ„õ†g‡g‰ىًKŒK޽޽�/‘/“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1·¾!1d·¾!1eÅ!1leÅ!1ïÎ!1èïÎ!1%Ò!1´%Ò!1ëç!1@ëç!1‰("1¼‰("17/"1Ä7/"1Á8"1@Á8"1÷;"1 ÷;"1½Q"1˜½Q"1[’"1[’"1 ™"1  ™"1“¢"1˜ “¢"1É¥"1d É¥"1 Ó"1ð  Ó"1aÚ"1Ì aÚ"1Áþ"1H Áþ"1,#1h,#1—:#1D—:#1So#1ÐSo#1c�#1ôc�#1aš#1 aš#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1·€&1à·€&1‡R'1$‡R'1/e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖxD b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G‡’ÒÎ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖxD b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G‡’ÒÎ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖxD  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G‡’ÒÎ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ·×�°Dd bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1a¡B¡á~¾ @€  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1daáBÂ#£„åeÆF§'ˆgçFÆÀ@ % ¥ y ù  „ 2 ² ã c ë k  B ¡ !  O € a¡B¡á@€ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPËË)*4¹4¹>0?0I¥I¥ST^•^•hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$_%_'Ö'Ö)M*M,Ä,Ä.;/;1‡1‡3þ3þ5u6u8ì8ì:c;c=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n1o1q¢q¢stv„v„xõxõzf{f}×}×H€H‚¹‚¹„*…*‡œ‡œ‰ Š Œ~Œ~ŽïŽï�`‘`“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1}}!1è}}!1¾!1d¾!1eÅ!1leÅ!1ïÎ!1èïÎ!1%Ò!1´%Ò!1Oç!1@Oç!1í'"1¼í'"17/"1Ä7/"1Á8"1@Á8"1÷;"1 ÷;"1!Q"1˜!Q"1¿‘"1¿‘"1 ™"1  ™"1“¢"1˜ “¢"1É¥"1d É¥"1 Ó"1ð  Ó"1mÚ"1Ì mÚ"1Íþ"1H Íþ"1,#1h,#1ß:#1Dß:#1›o#1Лo#1¹�#1ô¹�#1·›#1 ·›#1«$1¬«$1ƒò$1pƒò$1½Þ%1t½Þ%17ì%1T7ì%1{€&1à{€&1KR'1$KR'1ód'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ú¦ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  `G‡Î bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ú¦ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  `G‡Î bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{Ú¦  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  `G‡Î  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �°·×Pp bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1p° `8xÒ„Ä  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dpð`àPÐ@À0°  �‡€TÔòr!¡äd în Ö V È H º : ¬ , ž  I�€À0pâ"”Ô bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ Ô ÔII¾¾)3*34¨4¨>ñ>ñHŽIŽS T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÔÔJJÀÀ 6 6 ¬ ¬""˜˜„„ùùnnããX X"Í"Í$B%B'·'·),*,,¡,¡./1‹1‹346u6u8ê8ê:_;_=Ô=Ô?G@GBºBºD-E-G G IéIéK\L\NÏNÏPBQBSµSµURVRXÅXÅZ8[8]«]«_`b›b›deg�g�iôiôkglgnÚnÚpMqMsÀsÀu3v3x¦x¦z{}Œ}Œÿÿ�r‚r„å„å†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1 ~!1è ~!1«¾!1d«¾!1eÅ!1leÅ!1ïÎ!1èïÎ!1%Ò!1´%Ò!1ßç!1@ßç!1}("1¼}("17/"1Ä7/"1Á8"1@Á8"1÷;"1 ÷;"1±Q"1˜±Q"1O’"1O’"1 ™"1  ™"1“¢"1˜ “¢"1É¥"1d É¥"1 Ó"1ð  Ó"17Ú"1Ì 7Ú"1—þ"1H —þ"1Ù+#1hÙ+#17:#1D7:#1ón#1Ðón#1q�#1ôq�#1o˜#1 o˜#1É$1¬É$1¡ò$1p¡ò$1ÛÞ%1tÛÞ%1Uì%1TUì%1/�&1à/�&1ÿR'1$ÿR'1§e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•Éi5 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G‡™ÙÎ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•Éi5 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G‡™ÙÎ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•Éi5  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G‡™ÙÎ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ·×u•: bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1G‡ÎDU•Ü  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dGÇŽÕUœcã*ªñq¸8çgÿ¢"FÆ]Ý �˜ÔT ›  b â ) © ð p · ÷ ¸ ø > ~ Å LŒ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜTTÌÌ)D*D4¼4¼>4?4IªIªS T ^˜^˜hähär„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœŒŒ | |"ô"ô$l%l'ä'ä)\*\,Ô,Ô.L/L1Ä1Ä3<4<6´6´8,9,;¤;¤=>@”@”B C E„E„GüGüItJtLêLêN`O`QÖQÖSLTLVÂVÂX8Y8[®[®]$^$`š`šbce†e†gügüirjrlèlèn`o`qÖqÖsLtLvÂvÂx8y8{„{„}ú}úp€p‚æ‚æ„\…\‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1}}!1è}}!1¾!1d¾!1eÅ!1leÅ!1ïÎ!1èïÎ!1%Ò!1´%Ò!1Oç!1@Oç!1í'"1¼í'"17/"1Ä7/"1Á8"1@Á8"1÷;"1 ÷;"1!Q"1˜!Q"1¿‘"1¿‘"1 ™"1  ™"1“¢"1˜ “¢"1É¥"1d É¥"1 Ó"1ð  Ó"1UÚ"1Ì UÚ"1µþ"1H µþ"1÷+#1h÷+#1Ù:#1DÙ:#1•o#1Еo#1§�#1ô§�#1¥›#1 ¥›#1“$1¬“$1kò$1pkò$1¥Þ%1t¥Þ%1ì%1Tì%1€&1à€&1åQ'1$åQ'1�d'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ؤ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _G‡Î bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ؤ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _G‡Î bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{ؤ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _G‡Î  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ’²·×4T bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1q±"bÑÅ €À  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dqñbâSÓDÄ5µ&¦—ˆ÷wæf Õ U » ; Ä D ˆ  U Õ ³ 3 " ¢ ¢ " ‘ €–Öq±"bÑ€À bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ö ÖII¼¼)/*/4¢4¢>è>èH„I„SýSý]n^nhßhßrPsP} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††ùùllßßR R"Å"Å$8%8'«'«)*,‘,‘./1w1w3ê3ê5]6]8Ð8Ð:C;C=¶=¶?'@'B˜B˜D E GzGzIÀIÀK1L1N¢N¢PQS„S„U V X‘X‘Z[]s]s_ä_äa]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ü†üˆl‰l‹Ü‹Ü�LŽL�¼�¼’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1~!1è~!1±¾!1d±¾!1eÅ!1leÅ!1ïÎ!1èïÎ!1%Ò!1´%Ò!1åç!1@åç!1ƒ("1¼ƒ("17/"1Ä7/"1Á8"1@Á8"1÷;"1 ÷;"1·Q"1˜·Q"1U’"1U’"1 ™"1  ™"1“¢"1˜ “¢"1É¥"1d É¥"1 Ó"1ð  Ó"1gÚ"1Ì gÚ"1Çþ"1H Çþ"1 ,#1h ,#17:#1D7:#1ón#1Ðón#1q�#1ôq�#1o˜#1 o˜#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1·€&1à·€&1‡R'1$‡R'1/e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—Ëm9 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G‡˜ØÎ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—Ëm9 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G‡˜ØÎ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—Ëm9  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G‡˜ØÎ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ·×•µJj bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1G‡ÎU•¿ Ü  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dGÇŽÕUœcã*ªñq¸8ÿFÆ �ÔT ÕU ›  �  b â L Ì ) ©  ‡ ð p · ÷ > ~ 3 s Å LŒ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÄÄ):*:4°4°>&?&IšIšST^†^†hÏhÏrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$L%L'Â'Â)8*8,®,®.$/$1š1š346†6†8ü8ü:r;r=è=è?^@^BÔBÔDJEJGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqzqzsîsîubvbxÖxÖz{}“}“€‚{‚{„ï„ï†އމŠŒvŒvŽêŽê�^‘^“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ƒ}!1Ѓ}!1!¾!1L!¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Uç!1(Uç!1ó'"1¤ó'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1'Q"1€'Q"1Å‘"1üÅ‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#17#1,7#1ÿŒ#1˜ÿŒ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Û§ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TG‡Ó bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Û§ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TG‡Ó bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Û§  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TG‡Ó  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 “³·×Vv bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1'g€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ié4´ÿÊJ ×  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ù ÙNNÃÃ)8*84­4­>#?#I—I—S T ^^hóhórisi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A ¹ ¹11©©!!™™ƒƒøøm m"â"â$W%W'Ì'Ì)A*A,¶,¶.+/+1 1 346Š6Š8ÿ8ÿ:t;t=é=é?^@^BÓBÓDHEHG½G½I3J3L§L§NOQ�Q�STVwVwXëXëZ_[_]Ó]Ó_G`Gb»b»d/e/g£g£ijl‹l‹nÿnÿpsqssçsçu[v[xÏxÏzC{C}·}·+€+‚Ÿ‚Ÿ„…‡‰‡‰‰þ‰þ‹sŒsŽèŽè�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1~!1Ð~!1¥¾!1L¥¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Ùç!1(Ùç!1w("1¤w("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1«Q"1€«Q"1I’"1üI’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#1[7#1,[7#1÷‹#1˜÷‹#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1ù€&1Ìù€&1ÉR'1ÉR'1qe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1о^* b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M�Ø¥å bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1о^* b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M�Ø¥å bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1о^*  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M�Ø¥å  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ·×zš$D bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ÿ?€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ímª*gç$¤Ï   €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I£I£ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbèbèd]e]gÒgÒiGjGl¼l¼n1o1q¦q¦stv�v�xy{z{z}ï}ïd€d‚قلN…N‡ÇÉ8Š8Œ­Œ­Ž"�"‘—‘—“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1ƒ}!1Ѓ}!1!¾!1L!¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Uç!1(Uç!1ó'"1¤ó'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1'Q"1€'Q"1Å‘"1üÅ‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1 :#1, :#1‰�#1˜‰�#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1{€&1Ì{€&1KR'1KR'1ód'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ù¥ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UG‡Ò bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ù¥ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UG‡Ò bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ù¥  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UG‡Ò  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‘±·×2R bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1²ò €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1•  ` à + « ö v b¢ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×LLÁÁ)6*64«4«> ? I“I“ST^w^whéhér]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡üüqqææ[ ["Ð"Ð$E%E'º'º)/*/,¤,¤./1Ž1Ž346x6x8í8í:b;b=×=×?L@LBÁBÁD6E6G«G«I J L“L“NOQyQySìSìU_V_XÒXÒZE[E]¸]¸_+`+b�b�deg�g�ióiókelen×n×pIqIs»s»u-v-xŸxŸz{}ƒ}ƒõõ�g‚g„لنM‡M‰À‰À‹3Œ3ަަ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1 ~!1Ð ~!1«¾!1L«¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Ùç!1(Ùç!1w("1¤w("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1«Q"1€«Q"1I’"1üI’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1×9#1,×9#1uŒ#1˜uŒ#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀb. b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 NŽÚ¦æ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀb. b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 NŽÚ¦æ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀb.  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 NŽÚ¦æ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ·×šºPp bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1|¼ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1áa ž  [ Û  ˜ L Œ  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙOOÄÄ)9*94®4®>#?#I–I–S T ^|^|hïhïrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰ÿÿuuëëa a"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnänäpWqWsÊsÊu=v=x°x°z#{#}–}– € ‚|‚|„ï„ï†b‡b‰Õ‰Õ‹HŒH޻޻�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1‰}!1Љ}!1'¾!1L'¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1[ç!1([ç!1ù'"1¤ù'"1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1-Q"1€-Q"1Ë‘"1üË‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1s7#1,s7#1ÕŒ#1˜ÕŒ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1ç€&1Ìç€&1·R'1·R'1_e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ܨ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UI‰à  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ܨ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UI‰à  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ܨ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UI‰à   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ·×–¶Vv bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1V€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1NΗà`©) ¨ è  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>-?-I¤I¤ST^�^�hÿhÿrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜIUJULÌLÌNCOCQºQºS1T1V¨V¨XY[–[–] ^ `„`„bûbûdrergégéi`j`l×l×nIoIq»q»s-t-vŸvŸxy{ƒ{ƒ}õ}õg€g‚قلK…K‡½‡½‰/Š/Œ¡Œ¡Ž�‘…‘…“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1~!1Ð~!1Ÿ¾!1LŸ¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Óç!1(Óç!1q("1¤q("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1¥Q"1€¥Q"1C’"1üC’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1g7#1,g7#1uŒ#1˜uŒ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Û€&1ÌÛ€&1«R'1«R'1Se'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³S b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 X˜ŽÎî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³S b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 X˜ŽÎî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³S  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 X˜ŽÎî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ÿ.Na� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 J€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1(¨ëk®.qñJ Š  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh â âXXÎÎ)D*D4º4º>.?.I¢I¢ST^Š^Šhþhþrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²((žž Š Š"#%v%v'ì'ì)b*b,Ø,Ø.N/N1Ä1Ä3:4:6°6°8&9&;œ;œ=>@†@†BúBúDnEnGâGâIVJVLÊLÊN>O>Q²Q²S&T&VšVšXY[‚[‚]ö]ö_j`jbÞbÞdReRgÆgÆi:j:l®l®n"o"q–q–s t v~v~xòxòzf{f}Ú}ÚN€N‚‚„6…6‡ª‡ª‰ŠŒ�Œ�Ž�‘v‘v“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1‰}!1Љ}!1'¾!1L'¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Uç!1(Uç!1ó'"1¤ó'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1'Q"1€'Q"1Å‘"1üÅ‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1û9#1,û9#1Y�#1˜Y�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Ú¦ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 THˆÞ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Ú¦ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 THˆÞ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Ú¦  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 THˆÞ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �°·×0P bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ÿß €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1r ò ; »  „ Í M 1q €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ö ÖKKÀÀ)5*54ª4ª>!?!I–I–S T ^}^}hïhïrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÖÖKKÀÀ 5 5 ª ª””  ~~óóhhÝÝR R"Ç"Ç$<%<'±'±)&*&,›,›./1…1…3ú3ú5o6o8ä8ä:Y;Y=Î=Î?C@CB¸B¸D-E-G¢G¢IJLŽLŽNOQxQxSíSíUbVbX×X×ZL[L]Á]Á_6`6b«b«d e g•g•i j llnñnñpcqcsÕsÕuGvGx¹x¹z+{+}�}�€‚�‚�„ó„ó†e‡e‰׉׋IŒI޻޻�-‘-“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1~!1Ð~!1¥¾!1L¥¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Óç!1(Óç!1q("1¤q("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1¥Q"1€¥Q"1C’"1üC’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1é9#1,é9#1ùŒ#1˜ùŒ#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1c€&1Ìc€&13R'13R'1Ûd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µW# b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™�Íð0 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µW# b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™�Íð0 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µW#  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™�Íð0  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ÿ¿BbVv bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Í €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations14 ´ ÷ w º : } ý Í  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞRRÆÆ):*:4®4®> ? I’I’ST^v^vhèhèrZsZ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žž††úún n"â"â$V%V'Ê'Ê)>*>,²,².&/&1š1š346‚6‚8ö8ö:j;j=Þ=Þ?P@PBÂBÂD4E4G¦G¦IJLŠLŠNüNüPnQnSàSàURVRXÄXÄZ6[6]¨]¨_`bŒbŒdþdþfpgpiâiâkTlTnÆnÆp8q8sªsªuvxŽxŽz{}r}rää�V‚V„ȄȆ:‡:‰«‰«‹ŒŽ�Ž��þ�þ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1�}!1Ð�}!1-¾!1L-¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1aç!1(aç!1ÿ'"1¤ÿ'"1y/"1¬y/"19"1(9"19<"1ô9<"13Q"1€3Q"1Ñ‘"1üÑ‘"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#17#1,7#1;�#1˜;�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1ÿ€&1Ìÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ý© b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  I_Ÿ@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ý© b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  I_Ÿ@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ý©  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  I_Ÿ@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ·×™¹Xx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1\€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1pð@À�à` L €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee ß ßYYÏÏ)E*E4»4»>1?1I¥I¥ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯))££ — —"#%‡%‡'ý'ý)s*s,é,é._/_1Õ1Õ3K4K6Á6Á8797;­;­=#>#@™@™BCE…E…GûGûIqJqLåLåNYOYQÍQÍSATAVµVµX)Y)[�[�]^`…`…bùbùdmemgágáiUjUlÉlÉn=o=q±q±s%t%v™v™x y {�{�}õ}õi€i‚݂݄Q…Q‡Ňʼn9Š9Œ­Œ­Ž!�!‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1û}!1Ðû}!1™¾!1L™¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Çç!1(Çç!1e("1¤e("1[/"1¬[/"1å8"1(å8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#1…7#1,…7#1A�#1˜A�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1/�&1Ì/�&1ÿR'1ÿR'1§e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨H b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 X˜šÚâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨H b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 X˜šÚâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨H  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 X˜šÚâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 „¤8Xv– bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 `€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€NΜêj P €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh á áZZÓÓ)L*L4Å4Å>>?>IµIµS*T*^�^�hisƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhááZZÓÓ L L Å Å>>··00©©""›› � �"#%%'ø'ø)q*q,ê,ê.c/c1Ü1Ü3U4U6Î6Î8G9G;À;À=9>9@²@²B+C+E¤E¤GHJ–J–L M O„O„QûQûSrTrVéVéX`Y`[×[×]N^N`Å`Åb:c:e¯e¯g$h$j™j™lmo�o�qôqôsgtgvÚvÚxMyM{À{À}3~3€¦€¦‚ƒ…Œ…Œ‡ÿ‡ÿ‰qŠqŒãŒãŽU�U‘ǑǓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1�}!1Ð�}!1-¾!1L-¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1[ç!1([ç!1ù'"1¤ù'"1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1-Q"1€-Q"1Ë‘"1üË‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1[Ú"1´ [Ú"1»þ"10 »þ"1ý+#1Pý+#1 :#1, :#1Ë�#1˜Ë�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Û§ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  J^žþ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Û§ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  J^žþ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Û§  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  J^žþ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ·×�¯.N bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¬ì €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1° 0 € P Ð   œÜ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝUUÉÉ)=*=4±4±>%?%I—I—S T ^{^{híhír_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%���� } }"õ"õ$i%i'Ý'Ý)Q*Q,Å,Å.9/91­1­3!4!6•6•8 9 ;};}=ñ=ñ?e@eBÙBÙDMEMGÁGÁI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnãnãpUqUsÇsÇu9v9x«x«z{}�}�€‚s‚s„å„å†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1û}!1Ðû}!1™¾!1L™¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Íç!1(Íç!1k("1¤k("1[/"1¬[/"1å8"1(å8"1<"1ô<"1ŸQ"1€ŸQ"1=’"1ü=’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1:#1,:#1Ë�#1˜Ë�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªL b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™›Ûä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªL b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™›Ûä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªL  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™›Ûä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¤Äe…\| bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1®î €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¸ 8 †  T Ô " ¢ žÞ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ý ÝTTËË)B*B4¹4¹>0?0I¥I¥ST^‰^‰húhúrlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÝÝTTËË B B ¹ ¹00§§••  ƒƒúúq q"è"è$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n%o%q–q–stvxvxxéxézZ{Z}Ë}Ë<€<‚­‚­„…‡�‡�‰ŠŒrŒrŽãŽã�T‘T“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1•}!1Е}!13¾!1L3¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1aç!1(aç!1ÿ'"1¤ÿ'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"13Q"1€3Q"1Ñ‘"1üÑ‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1—7#1,—7#1Ž#1˜Ž#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÞª b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  JG‡Î bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÞª b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  JG‡Î bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÞª  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  JG‡Î  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 œ¼DdZz bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations12r€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÊJ¥%€ [ Û Ô €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØMMÂÂ)7*74¬4¬>?I’I’ST^x^xhëhër^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?K@KB¾B¾D1E1G¤G¤IJLŠLŠNýNýPpQpSãSãUVVVXÉXÉZ<[<]¯]¯_"`"b•b•deg{g{iîiîkalanÔnÔpGqGsºsºu-v-x x z{}†}†ùù�l‚l„߄߆R‡R‰ĉċ6Œ6ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1õ}!1Ðõ}!1“¾!1L“¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Áç!1(Áç!1_("1¤_("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1“Q"1€“Q"11’"1ü1’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#1�7#1,�7#1Ž#1˜Ž#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1A�&1ÌA�&1S'1S'1¹e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G‡Î�Í bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G‡Î�Í bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�=   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G‡Î�Í  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‰©Bbo� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations16v€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ØX±1Š c ã Ö €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àXXÐÐ)H*H4À4À>8?8I®I®S$T$^š^šhis†s†} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  �� € €"ø"ø$p%p'è'è)`*`,Ø,Ø.P/P1È1È3@4@6¸6¸8090;¨;¨= > @˜@˜BCEˆEˆGHJxJxLîLîNdOdQÚQÚSPTPVÆVÆX?IˆIˆSùSù]j^jhÛhÛrLsL} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠýýppããV V"É"É$<%<'¯'¯)"*",•,•./1{1{3î3î5a6a8Ô8Ô:G;G=º=º?+@+BœBœD E G~G~IïIïK`L`NÑNÑPBQBS³S³U$V$X•X•Z[]w]w_è_èaYbYdÊdÊf;g;i¬i¬klnŽnŽpÿpÿrpspuáuáwRxRzÃzÃ|4}4¥¥�‚„‡„‡†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1õ}!1Ðõ}!1“¾!1L“¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1Çç!1(Çç!1e("1¤e("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1™Q"1€™Q"17’"1ü7’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1�Ú"1´ �Ú"1ýþ"10 ýþ"1?,#1P?,#17:#1,7:#1£Ž#1˜£Ž#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G‡ÎŽÎ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G‡ÎŽÎ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G‡ÎŽÎ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ©ÉjŠb‚ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ï €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1< ¼  • î n Ç G o¯ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜRRÈÈ)>*>4´4´>*?*IžIžST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒxxîîd d"Ú"Ú$P%P'Æ'Æ)<*<,²,².(/(1ž1ž346Š6Š89;v;v=ì=ì?b@bBØBØDNENGÄGÄI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1›}!1Л}!19¾!1L9¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1gç!1(gç!1("1¤("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"19Q"1€9Q"1ב"1üב"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1µ7#1,µ7#1ÍŽ#1˜ÍŽ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€ß« b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @HˆÏ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€ß« b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @HˆÏ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€ß«  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @HˆÏ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ÿ¿Uu\| bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Hˆ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  †ìl Ò R ˜Ø €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝRRÇÇ)<*<4±4±>'?'I›I›ST^ƒ^ƒh÷h÷rmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��‡‡üüq q"æ"æ$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡�‡�‰ŠŒwŒwŽìŽì�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1é}!1Ðé}!1‡¾!1L‡¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1»ç!1(»ç!1Y("1¤Y("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1�Q"1€�Q"1+’"1ü+’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#1µ7#1,µ7#1ÓŽ#1˜ÓŽ#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’2þ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G‡Î‘Ñ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’2þ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G‡Î‘Ñ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’2þ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G‡Î‘Ñ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ŠªDdjŠ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1LŒ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations10°”øx Ü \ œÜ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>2?2I§I§ST^‘^‘his{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbìbìdaeagÖgÖiKjKlÀlÀn5o5qªqªstv”v”x y {~{~}ó}óh€h‚݂݄R…R‡LJlj<Š<Œ±Œ±Ž&�&‘›‘›“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1•}!1Е}!13¾!1L3¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1gç!1(gç!1("1¤("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"19Q"1€9Q"1ב"1üב"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#1[:#1,[:#1u�#1˜u�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Ý© b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @HˆÏ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Ý© b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @HˆÏ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€Ý©  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @HˆÏ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �­Nn*J bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1î. €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¸ 8 ž  „  j ê >~ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛPPÅÅ):*:4¯4¯>$?$I—I—S T ^{^{híhírasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹uuêê_ _"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I$J$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b¡b¡deg…g…i÷i÷kilinÛnÛpMqMs¿s¿u1v1x£x£z{}‡}‡ùù�k‚k„݄݆Q‡Q‰ĉċ7Œ7ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1ï}!1Ðï}!1�¾!1L�¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Áç!1(Áç!1_("1¤_("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#1[:#1,[:#1u�#1˜u�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Ÿ€&1ÌŸ€&1oR'1oR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”6 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G‡Î‘Ñ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”6 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G‡Î‘Ñ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”6  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G‡Î‘Ñ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¢Âi‰Tt bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ð0 €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1À @ ¤ $ ˆ  l ì @€ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝSSÈÈ)=*=4²4²>'?'IšIšS T ^€^€hóhórfsf} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��yyïïe e"Û"Û$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I+J+LžLžNOQ„Q„S÷S÷UjVjXÝXÝZP[P]Ã]Ã_6`6b©b©deg�g�ijlulunènèp[q[sÎsÎuAvAx´x´z'{'}š}š € ‚€‚€„ó„ó†f‡f‰ىًLŒLŽ¿Ž¿�2‘2“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1›}!1Л}!19¾!1L9¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1mç!1(mç!1 ("1¤ ("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1?Q"1€?Q"1Ý‘"1üÝ‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1Í7#1,Í7#1™�#1˜™�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1ÿ€&1Ìÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ଠb-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @HˆÐ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ଠb-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @HˆÐ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ଠ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @HˆÐ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ˜¸NnRr bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1^ž€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1xøiéZ Ú K Ë ^ž €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>1?1I¨I¨ST^‘^‘hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIYJYLÐLÐNGOGQ¾Q¾S5T5V¬V¬X#Y#[š[š]^`ˆ`ˆbÿbÿdvevgígíidjdlÛlÛnMoMq¿q¿s1t1v£v£xy{‡{‡}ù}ùk€k‚݂݄O…O‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1ã}!1Ðã}!1�¾!1L�¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1µç!1(µç!1S("1¤S("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1‡Q"1€‡Q"1%’"1ü%’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1mÚ"1´ mÚ"1Íþ"10 Íþ"1,#1P,#1Ó7#1,Ó7#1Ÿ�#1˜Ÿ�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1í€&1Ìí€&1½R'1½R'1ee'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡'ó b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G‡Î˜Ø bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡'ó b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G‡Î˜Ø bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡'ó  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G‡Î˜Ø  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ·×ƒ£6V bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1b¢€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ˆw÷f æ U Õ b¢ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll æ æ\\ÒÒ)H*H4¾4¾>2?2I¦I¦ST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllææ``ÚÚ T T Î ÎHHÂÂ<<¶¶,,¢¢ Ž Ž"#%z%z'ð'ð)f*f,Ü,Ü.R/R1È1È3>4>6´6´8*9*; ; =>@Š@ŠBþBþDrErGæGæIZJZLÎLÎNBOBQ¶Q¶S*T*VžVžXY[†[†]ú]ú_n`nbâbâdVeVgÊgÊi>j>l²l²n&o&qšqšstv‚v‚xöxözj{j}Þ}ÞR€R‚ƂƄ:…:‡®‡®‰!Š!Œ”Œ”Ž�‘z‘z“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1›}!1Л}!19¾!1L9¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1mç!1(mç!1 ("1¤ ("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1?Q"1€?Q"1Ý‘"1üÝ‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1:#1,:#1G�#1˜G�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1¥€&1Ì¥€&1uR'1uR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Þª b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @HˆÑ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Þª b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @HˆÑ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�Þª  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @HˆÑ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ŠªOo$D bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1O €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1< ¼ - ­  ž �O €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ú ÚOOÄÄ)9*94®4®>%?%IšIšST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÚÚOOÄÄ 9 9 ® ®##˜˜  ‚‚÷÷llááV V"Ë"Ë$@%@'µ'µ)***,Ÿ,Ÿ./1‰1‰3þ3þ5s6s8è8è:];]=Ò=Ò?G@GB¼B¼D1E1G¦G¦IJL’L’NOQ|Q|SñSñUfVfXÛXÛZP[P]Å]Å_:`:b¯b¯d$e$g™g™ijlƒlƒnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1é}!1Ðé}!1‡¾!1L‡¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1»ç!1(»ç!1Y("1¤Y("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1�Q"1€�Q"1+’"1ü+’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1:#1,:#1M�#1˜M�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1u€&1Ìu€&1ER'1ER'1íd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰+÷ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 G‡Î™Ù bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰+÷ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 G‡Î™Ù bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰+÷  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 G‡Î™Ù  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ·×—·>^ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Q €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1D Ä 3 ³ " ¢ ‘Q €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âVVÊÊ)>*>4²4²>$?$I–I–ST^z^zhìhìr^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢ŠŠþþr r"æ"æ$Z%Z'Î'Î)B*B,¶,¶.*/*1ž1ž346†6†8ú8ú:n;n=â=â?T@TBÆBÆD8E8GªGªIJLŽLŽNOQrQrSäSäUVVVXÈXÈZ:[:]¬]¬_`b�b�degtgtiæiækXlXnÊnÊp‡>‰¯‰¯‹ Œ ޑޑ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1¿}!1�¿}!1]¾!1]¾!1³Â!1$³Â!1‹ç!1�‹ç!1)("1)("1…,"1$…,"1]Q"1�]Q"1û‘"1û‘"1W–"1$W–"1Ú"1�Ú"1ßþ"1üßþ"1/0#1 /0#1û~#1| û~#1í$1Ü í$1Åò$1  Åò$1ÿÞ%1¤ ÿÞ%1#�&1„ #�&1óR'1ÈóR'1›e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1!U_“ò¾ýÉ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1!U_“ò¾ýÉ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1!U_“ò¾ýÉ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1:Z  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10zz ô ônnää)Z*Z4Ð4Ð>D?DI¸I¸S,T,^ ^ hisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðzzôônnèè b b Ü ÜVVÐÐJJÄÄ>>¸¸22 ¬ ¬"&#&%œ%œ'(*ˆ*ˆ,þ,þ.t/t1ê1ê3`4`6Ö6Ö8L9L;Â;Â=8>8@®@®B$C$EšEšGHJ„J„LøLøNlOlQàQàSTTTVÈVÈXI?IIÀIÀS5T5^¨^¨hisŒsŒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuîîggàà Y Y Ò ÒKKÄÄ==¶¶//¨¨!! š š"#%Œ%Œ'(*~*~,÷,÷.p/p1é1é3b4b6Û6Û8T9T;Í;Í=F>F@¿@¿B8C8E±E±G*H*J¡J¡LMO�O�QRT}T}VôVôXkYk[â[â]Y^Y`Ð`ÐbEcEeºeºg/h/j¤j¤lmoŒoŒqÿqÿsrtrvåvåxXyX{Ë{Ë}>~>€±€±‚$ƒ$…—…—‡ˆŠzŠzŒìŒìŽ^�^‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1¿}!1�¿}!1]¾!1]¾!1³Â!1$³Â!1‘ç!1�‘ç!1/("1/("1…,"1$…,"1cQ"1�cQ"1’"1’"1W–"1$W–"1…Ú"1�…Ú"1åþ"1üåþ"1/0#1 /0#1û~#1| û~#1Õ$1Ü Õ$1­ò$1  ­ò$1çÞ%1¤ çÞ%1«€&1„ «€&1{R'1È{R'1#e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1"V]‘îºÌ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1"V]‘îºÌ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1"V]‘îºÌ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Aa  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx ð ðhhÜÜ)P*P4Ä4Ä>6?6I¨I¨ST^Œ^Œhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðxxððhhàà X X Ð ÐHHÀÀ88°°((   � �"#%|%|'ð'ð)d*d,Ø,Ø.L/L1À1À34446¨6¨89;�;�=>@x@xBìBìD`E`GÔGÔIFJFL¸L¸N*O*QœQœSTV€V€XòXòZd[d]Ö]Ö_H`Hbºbºd,e,gžgžijl‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1ã}!1�ã}!1�¾!1 �¾!1YÅ!1YÅ!1µç!1€µç!1S("1üS("1+/"1+/"1‡Q"1p‡Q"1%’"1ì%’"1ý˜"1ôý˜"1ùØ"1`ùØ"1Yý"1ÌYý"1/0#1ì/0#1û~#1L û~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{é b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¦æ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{é b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¦æ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1G{é b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¦æ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1·×  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu ì ìccÚÚ)Q*Q4È4È>=?=I²I²S%T%^–^–hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuììccÚÚ Q Q È È??¶¶--¤¤’’  € €"÷"÷$n%n'å'å)\*\,Ó,Ó.J/J1Á1Á38486¯6¯8&9&;�;�=>@‹@‹BCEyEyGðGðIeJeLÚLÚNOOOQÄQÄS9T9V®V®X#Y#[˜[˜] ^ `‚`‚bõbõdhehgÛgÛiNjNlÁlÁn2o2q£q£stv…v…xöxözg{g}Ø}ØI€I‚º‚º„+…+‡›‡›‰ Š Œ}Œ}ŽîŽî�_‘_“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1³}!1�³}!1Q¾!1 Q¾!1³Â!1³Â!1ç!1€ç!1("1ü("1…,"1…,"1QQ"1pQQ"1ï‘"1ìï‘"1W–"1ôW–"1ùØ"1`ùØ"1Yý"1ÌYý"1/0#1ì/0#1û~#1L û~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ï» b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ï» b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ï» b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1·×  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss é é^^ÓÓ)H*H4»4»>.?.I¡I¡ST^‡^‡húhúrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðsséé__ÕÕ K K Á Á77­­##™™ƒƒøøm m"â"â$W%W'Ì'Ì)A*A,¶,¶.+/+1 1 346Š6Š8ÿ8ÿ:t;t=ç=ç?Z@ZBÍBÍD@E@G³G³I&J&L™L™N O QQSòSòUeVeXØXØZK[K]¾]¾_1`1b¤b¤degŠgŠiýiýkplpnãnãpVqVsÉsÉu???IµIµS+T+^¡^¡his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqééaaÙÙ Q Q É ÉAA¹¹11©©!!™™ ‰ ‰"#%y%y'ñ'ñ)i*i,á,á.Y/Y1Ñ1Ñ3I4I6Á6Á8999;±;±=)>)@¡@¡BCE‘E‘G H JJLõLõNkOkQáQáSWTWVÍVÍXCYC[¹[¹]/^/`¥`¥bce‘e‘ghj}j}lólónioiqßqßsUtUvËvËxAyA{·{·}-~-€£€£‚ƒ…�…�‡ˆŠ{Š{ŒñŒñŽg�g‘ݑݓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1­}!1�­}!1K¾!1 K¾!1³Â!1³Â!1ç!1€ç!1("1ü("1…,"1…,"1QQ"1pQQ"1ï‘"1ìï‘"1W–"1ôW–"1ùØ"1`ùØ"1Yý"1ÌYý"1/0#1ì/0#1û~#1L û~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%19€&1T 9€&1 R'1˜ R'1±d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽë· b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽë· b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽë· b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1·×  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss é é\\ÏÏ)B*B4³4³>$?$I•I•ST^w^whèhèrWsW} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðsséé__ÕÕ K K Á Á77­­##™™  òòe e"Ø"Ø$K%K'¾'¾)1*1,¤,¤./1Š1Š3ý3ý5p6p8ã8ã:V;V=Ç=Ç?8@8B©B©DEG‹G‹IüIüKmLmNÞNÞPOQOSÀSÀU1V1X¢X¢Z[]„]„_õ_õafbfd×d×fHgHi¹i¹k*l*n›n›p q s}s}uîuîw_x_zÐzÐ|A}A²²�#‚#„”„”†‡‰s‰s‹ã‹ã�SŽS�Ã�Ã’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1é}!1�é}!1‡¾!1 ‡¾!1GÅ!1GÅ!1µç!1€µç!1S("1üS("1/"1/"1‡Q"1p‡Q"1%’"1ì%’"1ë˜"1ôë˜"1ùØ"1`ùØ"1Yý"1ÌYý"1/0#1ì/0#1û~#1L û~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1á€&1T á€&1±R'1˜±R'1Ye'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„&ò b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1™Ù€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„&ò b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1™Ù€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„&ò b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1™Ù €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1·×  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ç ç]]ÓÓ)I*I4¿4¿>3?3I§I§ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqçç]]ÓÓ I I ¿ ¿55««!!——  ƒƒùùo o"å"å$[%['Ñ'Ñ)G*G,½,½.3/31©1©346•6•8 9 ;�;�=÷=÷?m@mBãBãDYEYGÏGÏICJCL·L·N+O+QŸQŸSTV‡V‡XûXûZo[o]ã]ã_W`WbËbËd?e?g³g³i'j'l›l›noqƒqƒs÷s÷ukvkxßxßzS{S}Ç}Ç;€;‚¯‚¯„#…#‡—‡—‰ Š ŒŒŽóŽó�g‘g“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1§}!1�§}!1E¾!1 E¾!1³Â!1³Â!1yç!1€yç!1("1ü("1…,"1…,"1KQ"1pKQ"1é‘"1ìé‘"1W–"1ôW–"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ì¸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ì¸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ì¸ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1·×  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ê ê__ÔÔ)I*I4¾4¾>2?2I¦I¦ST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrêêbbÚÚ R R Ê ÊBBºº22ªª””  ~ ~"ó"ó$h%h'Ý'Ý)R*R,Ç,Ç.e>g²g²i&j&lšlšnoq‚q‚sösöujvjxÞxÞzR{R}Æ}Æ:€:‚®‚®„"…"‡–‡–‰ Š Œ€Œ€ŽõŽõ�j‘j“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1é}!1�é}!1‡¾!1 ‡¾!1SÅ!1SÅ!1»ç!1€»ç!1Y("1üY("1%/"1%/"1�Q"1p�Q"1+’"1ì+’"1÷˜"1ô÷˜"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�-ù b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¥å€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�-ù b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¥å€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�-ù b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¥å €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1·×  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ä ä[[ÒÒ)I*I4À4À>5?5IªIªST^”^”h i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmää[[ÒÒ I I À À77®®%%œœŠŠ x x"ï"ï$f%f'Ý'Ý)T*T,Ë,Ë.B/B1¹1¹30406§6§89;•;•= > @ƒ@ƒBúBúDqEqGèGèI]J]LÒLÒNGOGQ¼Q¼S1T1V¦V¦XY[�[�]^`z`zbïbïddedgÙgÙiNjNlÃlÃn8o8q­q­s"t"v—v—x y {�{�}ö}ök€k‚à‚à„U…U‡ʇʉ?Š?Œ´Œ´Ž)�)‘ž‘ž“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1§}!1�§}!1E¾!1 E¾!1¹Â!1¹Â!1sç!1€sç!1("1ü("1‹,"1‹,"1EQ"1pEQ"1ã‘"1ìã‘"1]–"1ô]–"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹è´ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹è´ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹è´ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1·×  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æ[[ÐÐ)E*E4º4º>-?-I I ST^„^„höhörhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  €€õõj j"ß"ß$T%T'É'É)>*>,³,³.(/(1�1�346‡6‡8ü8ü:q;q=æ=æ?[@[BÐBÐDEEEGºGºI-J-L L NOQ†Q†SùSùUlVlXßXßZR[R]Å]Å_8`8bªbªdegŽgŽijlrlrnänäpVqVsÈsÈu:v:x¬x¬z{}�}�€‚t‚t„æ„æ†X‡X‰ˉˋ>Œ>ޱޱ�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1ï}!1�ï}!1�¾!1 �¾!1SÅ!1SÅ!1»ç!1€»ç!1Y("1üY("1%/"1%/"1�Q"1p�Q"1+’"1ì+’"1÷˜"1ô÷˜"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�/û b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¤ä€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�/û b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¤ä€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�/û b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¤ä €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1·×  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn ä äZZÏÏ)D*D4¹4¹>,?,IŸIŸST^…^…høhørksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnääZZÐÐ F F ¼ ¼22¨¨””  €€ööl l"â"â$W%W'Ì'Ì)A*A,¶,¶.+/+1 1 346Š6Š8ÿ8ÿ:t;t=é=é?^@^BÓBÓDHEHG½G½I0J0L£L£NOQ‰Q‰SüSüUoVoXâXâZU[U]È]È_;`;b®b®d!e!g”g”ijlzlzníníp`q`sÓsÓuFvFx¹x¹z,{,}Ÿ}Ÿ€‚…‚…„ø„ø†k‡k‰Þ‰Þ‹QŒQŽÄŽÄ�7‘7“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¡}!1�¡}!1?¾!1 ?¾!1¿Â!1¿Â!1sç!1€sç!1("1ü("1‘,"1‘,"1EQ"1pEQ"1ã‘"1ìã‘"1c–"1ôc–"1íØ"1`íØ"1Mý"1ÌMý"1#0#1ì#0#1ï~#1L ï~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠéµ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1V€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠéµ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1V€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠéµ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1V €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1·×  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn å å\\ÓÓ)J*J4Á4Á>8?8I¯I¯S&T&^˜^˜h i s|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnåå\\ÓÓ J J Á Á88¯¯&&��‹‹ y y"ð"ð$g%g'Þ'Þ)U*U,Ì,Ì.C/C1º1º31416¨6¨89;–;–= > @„@„BûBûDrErGéGéI`J`L×L×NNONQÅQÅS1?1I¥I¥ST^�^�hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmççaaÛÛ U U Ï ÏIIÃÃ==··--££ � �"#%{%{'ñ'ñ)g*g,Ý,Ý.S/S1É1É3?4?6µ6µ8+9+;¡;¡=>@‰@‰BýBýDqEqGåGåIYJYLÍLÍNAOAQµQµS)T)V�V�XY[…[…]ù]ù_m`mbábádUeUgÉgÉi=j=l±l±n%o%q™q™s t v�v�xõxõzi{i}Ý}ÝQ€Q‚łń9…9‡«‡«‰ŠŒ‘Œ‘Ž�‘w‘w“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¡}!1�¡}!1?¾!1 ?¾!1¿Â!1¿Â!1mç!1€mç!1 ("1ü ("1‘,"1‘,"1?Q"1p?Q"1Ý‘"1ìÝ‘"1c–"1ôc–"1íØ"1`íØ"1Mý"1ÌMý"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆå± b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1W€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆå± b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1W€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆå± b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1W €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1·×  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll á áVVËË)@*@4µ4µ>*?*IŸIŸST^†^†høhørjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllááVVËË @ @ µ µ**ŸŸ‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?N@NBÃBÃD8E8G­G­I"J"L—L—N O Q�Q�SöSöUkVkXàXàZU[U]Ê]Ê_?`?b´b´d)e)gžgžijlˆlˆnúnúplqlsÞsÞuPvPxÂxÂz4{4}¦}¦€‚ЂЄü„ü†n‡n‰à‰à‹RŒRŽÄŽÄ�6‘6“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1ï}!1�ï}!1�¾!1 �¾!1;Å!1;Å!1Áç!1€Áç!1_("1ü_("1 /"1 /"1“Q"1p“Q"11’"1ì1’"1ߘ"1ôߘ"1�Ú"1`�Ú"1ýþ"1Ìýþ"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1o€&1T o€&1?R'1˜?R'1çd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–8 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ѐ} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–8 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ѐ} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–8 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ð €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1e…  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm å åYYÍÍ)A*A4³4³>%?%I—I—S T ^{^{híhír]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmåå]]ÕÕ M M Å Å==µµ--¥¥�� u u"é"é$]%]'Ñ'Ñ)E*E,¹,¹.-/-1¡1¡346‰6‰8ý8ý:q;q=ã=ã?U@UBÇBÇD9E9G«G«IJL�L�NOQsQsSåSåUWVWXÉXÉZ;[;]­]­_`b‘b‘deguguiçiçkYlYnËnËp=q=s¯s¯u!v!x“x“z{}w}wéé�[‚[„̈́͆=‡=‰®‰®‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1›}!1�›}!19¾!1 9¾!1³Â!1³Â!1mç!1€mç!1 ("1ü ("1…,"1…,"1?Q"1p?Q"1Ý‘"1ìÝ‘"1W–"1ôW–"1íØ"1`íØ"1Mý"1ÌMý"1#0#1ì#0#1ï~#1L ï~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡æ² b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡æ² b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡æ² b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1·×  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn è èbbØØ)N*N4Ä4Ä>8?8I¬I¬S T ^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnèèbbÜÜ V V Ð ÐJJÄÄ>>¸¸22¬¬&&    "#%�%�'(*|*|,ò,ò.h/h1Þ1Þ3T4T6Ê6Ê8@9@;¶;¶=,>,@¢@¢BCEŽEŽGHJxJxLìLìN`O`QÔQÔSHTHV¼V¼X0Y0[¤[¤]^`Œ`Œbcetetgègèi\j\lÐlÐnDoDq¸q¸s,t,v v xy{ˆ{ˆ}ü}üp€p‚ä‚ä„X…X‡̇̉@Š@Œ´Œ´Ž(�(‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1õ}!1�õ}!1“¾!1 “¾!1GÅ!1GÅ!1Çç!1€Çç!1e("1üe("1/"1/"1™Q"1p™Q"17’"1ì7’"1ë˜"1ôë˜"1‘Ú"1`‘Ú"1ñþ"1Ìñþ"1#0#1ì#0#1ï~#1L ï~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1/�&1T /�&1ÿR'1˜ÿR'1§e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸ?  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¢â€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸ?  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¢â€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸ?  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¢â €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1[{  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii â â[[ÔÔ)M*M4Æ4Æ>=?=I´I´S)T)^œ^œhis€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiââ[[ÔÔ M M Æ Æ??¸¸11ªª##œœ Ž Ž"#%€%€'ù'ù)r*r,ë,ë.d/d1Ý1Ý3V4V6Ï6Ï8H9H;Á;Á=:>:@³@³B,C,E¥E¥GHJ•J•L M OƒOƒQúQúSqTqVèVèX_Y_[Ö[Ö]M^M`Ä`Äb9c9e®e®g#h#j˜j˜l m o€o€qóqósftfvÙvÙxLyL{¿{¿}2~2€¥€¥‚ƒ…‹…‹‡ü‡ü‰nŠnŒàŒàŽR�R‘đē b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1•}!1�•}!13¾!1 3¾!1³Â!1³Â!1gç!1€gç!1("1ü("1…,"1…,"19Q"1p9Q"1ב"1ìב"1W–"1ôW–"1íØ"1`íØ"1Mý"1ÌMý"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…â® b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…â® b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…â® b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1·×  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ä ä\\ÐÐ)D*D4¸4¸>*?*IœIœST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllää\\ÔÔ L L Ä Ä<<´´,,¤¤””  „ „"ü"ü$p%p'ä'ä)X*X,Ì,Ì.@/@1´1´3(4(6œ6œ89;„;„=ø=ø?l@lBàBàDTETGÈGÈI:J:L¬L¬NOQ�Q�STVtVtXæXæZX[X]Ê]Ê_<`v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1õ}!1�õ}!1“¾!1 “¾!1GÅ!1GÅ!1Çç!1€Çç!1e("1üe("1/"1/"1™Q"1p™Q"17’"1ì7’"1ë˜"1ôë˜"1£Ú"1`£Ú"1ÿ"1Ìÿ"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1½€&1T ½€&1�R'1˜�R'15e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¡á€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¡á€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¡á €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1r’  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii à àWWÎÎ)E*E4¼4¼>1?1I¦I¦ST^Š^Šhûhûrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiààWWÎÎ E E ¼ ¼33ªª!!˜˜††ýýt t"ë"ë$b%b'Ù'Ù)P*P,Ç,Ç.>/>1µ1µ3,4,6£6£89;‘;‘=>@@BöBöDmEmGäGäIYJYLÎLÎNCOCQ¸Q¸S-T-V¢V¢XY[Œ[Œ]^`v`vbébéd\e\gÏgÏiBjBlµlµn&o&q—q—stvyvyxêxêz[{[}Ì}Ì=€=‚®‚®„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1•}!1�•}!13¾!1 3¾!1³Â!1³Â!1gç!1€gç!1("1ü("1…,"1…,"19Q"1p9Q"1ב"1ìב"1W–"1ôW–"1IÚ"1`IÚ"1©þ"1Ì©þ"10#1ì0#1é~#1L é~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„㯠b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1S€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„㯠b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1S€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„㯠b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1S €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1;  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝRRÇÇ)<*<4¯4¯>"?"I•I•ST^{^{hîhîr_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��wwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Û=Û?N@NBÁBÁD4E4G§G§IJL�L�NOQsQsSæSæUYVYXÌXÌZ?[?]²]²_%`%b˜b˜d e g~g~iñiñkdldn×n×pJqJs½s½u0v0x£x£z{}‰}‰üü�o‚o„â„â†S‡S‰ʼnŋ7Œ7ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1û}!1�û}!1™¾!1 ™¾!1/Å!1/Å!1Çç!1€Çç!1e("1üe("1/"1/"1™Q"1p™Q"17’"1ì7’"1Ó˜"1ôÓ˜"1‘Ú"1`‘Ú"1ñþ"1Ìñþ"10#1ì0#1é~#1L é~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1;�&1T ;�&1 S'1˜ S'1³e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨H b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ŒÌ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨H b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ŒÌ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨H b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ŒÌ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1`€  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝUUÍÍ)E*E4½4½>3?3I©I©ST^•^•h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%���� } }"õ"õ$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=>@•@•B C E…E…GýGýIsJsLéLéN_O_QÕQÕSKTKVÁVÁX7Y7[­[­]#^#`™`™bce…e…gûgûiqjqlçlçn]o]qÓqÓsItIv¿v¿x5y5{«{«}!~!€—€—‚ ƒ …ƒ…ƒ‡ù‡ù‰oŠoŒåŒåŽ[�[‘ёѓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1�}!1��}!1-¾!1 -¾!1³Â!1³Â!1aç!1€aç!1ÿ'"1üÿ'"1…,"1…,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1W–"1ôW–"1aÚ"1`aÚ"1Áþ"1ÌÁþ"10#1ì0#1é~#1L é~#1™$1¬ ™$1qò$1p qò$1«Þ%1t «Þ%1!€&1T !€&1ñQ'1˜ñQ'1™d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ß« b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1T€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ß« b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1T€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ß« b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1T €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations10P  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝPPÃÃ)6*64§4§>?I‰I‰SúSú]k^khÜhÜrKsK} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��ssææY Y"Ì"Ì$?%?'²'²)%*%,˜,˜. / 1~1~3ñ3ñ5d6d8×8×:J;J=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éaZbZdËdËf'?'I›I›ST^ƒ^ƒh÷h÷rksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡‹‡‹‰ÿ‰ÿ‹sŒsŽçŽç�[‘[“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1�}!1��}!1-¾!1 -¾!1­Â!1­Â!1[ç!1€[ç!1ù'"1üù'"1,"1,"1-Q"1p-Q"1Ë‘"1ìË‘"1Q–"1ôQ–"1IÚ"1`IÚ"1©þ"1Ì©þ"10#1ì0#1ã~#1L ã~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1À&1T À&1“R'1˜“R'1;e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ଠb-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ଠb-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ଠb-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1 @  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞSSÈÈ)=*=4²4²>&?&IšIšST^„^„høhørlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžˆˆýýr r"ç"ç$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`JbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡Œ‡Œ‰ŠŒvŒvŽëŽë�`‘`“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1~!1�~!1Ÿ¾!1 Ÿ¾!1;Å!1;Å!1Íç!1€Íç!1k("1ük("1 /"1 /"1ŸQ"1pŸQ"1=’"1ì=’"1ߘ"1ôߘ"1aÚ"1`aÚ"1Áþ"1ÌÁþ"10#1ì0#1ã~#1L ã~#1É$1¬ É$1ñð$1p ñð$1ÏÀ%1t ÏÀ%1ÿ€&1T ÿ€&1õP'1˜õP'1�c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±Q b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ÿß€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±Q b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ÿß€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±Q b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ÿß €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1YU  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>)?)IžIžST^’^’his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜIQJQLÆLÆN;O;Q°Q°S%T%VšVšXY[„[„]ù]ù_n`nbãbãdXeXgÍgÍiBjBl·l·n6o6q«q«s t v•v•x y {{}ô}ôi€i‚Þ‚Þ„S…S‡ȇȉ=Š=Œ²Œ²Ž'�'‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1‰}!1�‰}!1'¾!1 '¾!1§Â!1§Â!1[ç!1€[ç!1ù'"1üù'"1y,"1y,"1-Q"1p-Q"1Ë‘"1ìË‘"1K–"1ôK–"1gÚ"1`gÚ"1Çþ"1ÌÇþ"10#1ì0#1ã~#1L ã~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kܨ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kܨ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kܨ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1=]  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94®4®>!?!I”I”S T ^|^|hîhîr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?O@OBÄBÄD9E9G®G®I!J!L”L”NOQzQzSíSíU`V`XÓXÓZF[F]¹]¹_,`,b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†P‡P‰ÉË6Œ6ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1~!1�~!1Ÿ¾!1 Ÿ¾!1;Å!1;Å!1Óç!1€Óç!1q("1üq("1 /"1 /"1¥Q"1p¥Q"1C’"1ìC’"1ߘ"1ôߘ"1‹Ú"1`‹Ú"1ëþ"1Ìëþ"10#1ì0#1ã~#1L ã~#1±$1¬ ±$1Ùð$1p Ùð$1·À%1t ·À%1™€&1T ™€&1�P'1˜�P'17c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±S b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1žÞ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±S b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1žÞ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±S b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1žÞ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1†‚  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÃÃ)8*84­4­> ? I“I“ST^�^�hôhôrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD.?.I¥I¥ST^š^šh i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLÍLÍNDODQ»Q»S2T2V©V©X Y [—[—]^`…`…bübüdsesgêgêiajalØlØnVoVqÈqÈs:t:v¬v¬xy{�{�}~€t€t‚æ‚æ„X…X‡ʇʉ<Š<Œ®Œ®Ž � ‘’‘’“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1~!1�~!1Ÿ¾!1 Ÿ¾!1#Å!1#Å!1Óç!1€Óç!1q("1üq("1õ."1õ."1¥Q"1p¥Q"1C’"1ìC’"1ǘ"1ôǘ"1Ú"1`Ú"1þ"1Ôþ"10#1ô0#1ã~#1T ã~#1Ã$1´ Ã$1ëð$1x ëð$1ÉÀ%1| ÉÀ%1á€&1\ á€&1×P'1 ×P'1c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºZ& b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‰É€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºZ& b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‰É€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºZ& b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‰É €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1£ëc  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Û ÛQQÑÑ)G*G4»4»>/?/I£I£S#T#^—^—h i s}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÛÛUUÏÏ I I à Ã==··11««!!——  ƒ ƒ"ù"ù$y%y'ï'ï)e*e,Û,Û.Q/Q1Ç1Ç3=4=6³6³8)9);Ÿ;Ÿ=>@‡@‡BûBûDoEoGãGãIWJWLËLËN?O?Q³Q³S'T'V›V›XY[ƒ[ƒ]÷]÷_k`kbëbëd_e_gÓgÓiGjGl»l»n/o/q£q£stv‹v‹xÿxÿzs{s}ç}ç[€[‚ςτC…C‡µ‡µ‰(Š(Œ›Œ›Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1ƒ}!1�ƒ}!1!¾!1 !¾!1§Â!1§Â!1Uç!1€Uç!1ó'"1üó'"1y,"1y,"1'Q"1p'Q"1Å‘"1ìÅ‘"1K–"1ôK–"1½Ø"1`½Ø"1ý"1Ìý"10#1ì0#1ã~#1L ã~#1·$1¬ ·$1ßð$1p ßð$1½À%1t ½À%1™€&1T ™€&1�P'1˜�P'17c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ù¥ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ù¥ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1H|Ù¥ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1·³  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Õ ÕJJÃÃ)8*84­4­>"?"I—I—S T ^Š^Šhühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÕÕJJ¿¿ 4 4 © ©““}}òòggÜÜQ Q"Æ"Æ$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nþnþppqpsâsâuTvTxÆxÆz8{8}ª}ª€‚ނބ…‡r‡r‰ä‰ä‹VŒVŽÈŽÈ�:‘:“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1~!1�~!1¥¾!1 ¥¾!1#Å!1#Å!1Óç!1€Óç!1q("1üq("1õ."1õ."1¥Q"1p¥Q"1C’"1ìC’"1ǘ"1ôǘ"1Ú"1`Ú"1þ"1Ôþ"10#1ô0#1ã~#1T ã~#1«$1´ «$1Óð$1x Óð$1±À%1| ±À%1i€&1\ i€&1_P'1 _P'1c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†º\( b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ŠÊ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†º\( b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ŠÊ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†º\( b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ŠÊ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1£ëa  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ù ÙMMÉÉ)=*=4¯4¯>!?!I“I“ST^ƒ^ƒhõhõrese} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A ¹ ¹11©©!!™™  ��õõi i"Ý"Ý$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b™b™d e g}g}iïiïkalanÓnÓpEqEs·s·u)v)x›x›z { }}ññ�c‚c„Õ„Õ†E‡E‰¶‰¶‹'Œ'Ž˜Ž˜� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„%i!1M}!1M}!1ë½!1|ë½!1OÃ!1„OÃ!1…Æ!1ð…Æ!1Ð!1|Ð!1EÓ!1HEÓ!1ç!1Ôç!1½'"1P½'"1!-"1X!-"1W0"1ÄW0"1á9"1Pá9"1="1="1ñP"1¨ñP"1�‘"1$ �‘"1ó–"1, ó–"1)š"1˜ )š"1³£"1$ ³£"1é¦"1ð é¦"1+Ô"1| +Ô"1yÚ"1XyÚ"1Ùþ"1ÔÙþ"1,#1ô,#1=:#1Ð=:#1ùn#1\ùn#1›�#1€›�#1™˜#1¬™˜#1Õ$18Õ$1­ò$1ü­ò$1çÞ%1çÞ%1aì%1àaì%1;�&1l;�&1 S'1° S'1³e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P±} b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¹ù bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P±} b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¹ù  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P±}  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¹ù bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 q‘/Vv bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1OI‰Ò[›ä$  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d�IÉÎN’� Û[L̤$mí6¶ÿÈH‘ZÚ#£ìl µ 5 ~ þ G Ç  � ÿ ? Ù  b ¢ ë +t´ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ « «OOÑÑ)I*I4Á4Á>9?9I¯I¯S%T%^§^§his“s“} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__××OOÇÇ ? ? ‹ ‹{{óókk‡‡ÿÿw w"ï"ï$q%q'é'é)a*a,Ù,Ù.Q/Q1É1É3A4A6¹6¹8191;©;©=!>!@™@™BCE‰E‰GHJyJyLïLïNeOeQÛQÛSQTQVÇVÇX=Y=[³[³])^)`Ÿ`Ÿbce‹e‹ghjwjwlílínoooqåqås[t[vÑvÑxGyG{½{½}3~3€©€©‚ƒ…•…•‡ ˆ Š�Š�Œ÷Œ÷Žm�m‘ã‘ã“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1·¾!1d·¾!1kÅ!1lkÅ!1õÎ!1èõÎ!1+Ò!1´+Ò!1ëç!1@ëç!1‰("1¼‰("1=/"1Ä=/"1Ç8"1@Ç8"1ý;"1 ý;"1½Q"1˜½Q"1[’"1[’"1™"1 ™"1™¢"1˜ ™¢"1Ï¥"1d Ï¥"1Ó"1ð Ó"1UÚ"1Ì UÚ"1µþ"1H µþ"1÷+#1h÷+#1=:#1D=:#1ùn#1Ðùn#1›�#1ô›�#1™˜#1 ™˜#1Ã$1¬Ã$1›ò$1p›ò$1ÕÞ%1tÕÞ%1Oì%1TOì%1í€&1àí€&1½R'1$½R'1ee'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖxD b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I‰�ÍÒ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖxD b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I‰�ÍÒ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖxD  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I‰�ÍÒ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¹Ù‡§:Z bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1QI‰Ò[›ä$  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d‘IÉÊJ”ƒÝ]<¼¦&oï8¸�ÊJ“\Ü%¥în · 7 € I É ¢ â  ’ Û  d ¤ í -v¶ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÉÉ)*4·4·>0?0I§I§ST^œ^œhis€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$]%]'Ô'Ô)K*K,Â,Â.9/91„1„3û3û5r6r8é8é:`;`=>@z@zBñBñDhEhGßGßIXJXLÏLÏNFOFQ½Q½S4T4V«V«X"Y"[™[™]^`‡`‡bþbþdueugìgìicjclÚlÚnXoXqÊqÊsŠ>Œ°Œ°Ž"�"‘”‘”“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„%i!1M}!1M}!1ë½!1|ë½!1OÃ!1„OÃ!1…Æ!1ð…Æ!1Ð!1|Ð!1EÓ!1HEÓ!1ç!1Ôç!1·'"1P·'"1!-"1X!-"1W0"1ÄW0"1á9"1Pá9"1="1="1ëP"1¨ëP"1‰‘"1$ ‰‘"1ó–"1, ó–"1)š"1˜ )š"1³£"1$ ³£"1é¦"1ð é¦"1+Ô"1| +Ô"1gÚ"1XgÚ"1Çþ"1ÔÇþ"1 ,#1ô ,#1y:#1Ðy:#15o#1\5o#1ÇŽ#1€ÇŽ#1Å™#1¬Å™#1½$18½$1•ò$1ü•ò$1ÏÞ%1ÏÞ%1Iì%1àIì%1À&1lÀ&1“R'1°“R'1;e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O®z b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¹ù bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O®z b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¹ù  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O®z  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¹ù bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ž®%ELl bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Y™ŽÎò2‰É `  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dYÙ2² ‹äd ‹½=ÊJ–‰ oïHÈHÈŸövÍM ¤ $ { û R Ò ) © ~ ¾  € Ù r² I à bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ ¨ ¨KKÉÉ)?*?4µ4µ>+?+IŸIŸST^“^“his{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 € €ööllââXXûûqqçç] ]"Ó"Ó$Q%Q'Ç'Ç)=*=,³,³.)/)1Ÿ1Ÿ346‹6‹89;w;w=í=í?c@cBÙBÙDOEOGÅGÅI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÃbÃd7e7g«g«ijl“l“noq‡q‡sûsûuovoxãxãzW{W}Ë}Ë?€?‚³‚³„'…'‡›‡›‰ŠŒƒŒƒŽ÷Ž÷�k‘k“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx%i!1~!1è~!1½¾!1d½¾!1kÅ!1lkÅ!1õÎ!1èõÎ!1+Ò!1´+Ò!1ñç!1@ñç!1�("1¼�("1=/"1Ä=/"1Ç8"1@Ç8"1ý;"1 ý;"1ÃQ"1˜ÃQ"1a’"1a’"1™"1 ™"1™¢"1˜ ™¢"1Ï¥"1d Ï¥"1Ó"1ð Ó"11Ú"1Ì 1Ú"1‘þ"1H ‘þ"1Ó+#1hÓ+#1m:#1Dm:#1)o#1Ð)o#1‹Ž#1ô‹Ž#1‰™#1 ‰™#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1“€&1à“€&1cR'1$cR'1 e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¤Ø|H b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I‰ŽÎÒ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¤Ø|H b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I‰ŽÎÒ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¤Ø|H  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I‰ŽÎÒ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¹Ùƒ£6 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1V–ŠÊì,€ÀT  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dVÖ,¬‚ØXõu®.¯/„hèZÚ!¡0°„ØX¬, € T Ô ( ¨ ü |  [ Ð P ¦ æ <|Ðd¤ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Õ ÕJJÃÃ) * 4­4­>$?$I™I™ST^Œ^Œhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÕÕJJ¿¿ 4 4 © ©““}}òòggÜÜQ Q"Æ"Æ$?%?'´'´))*),ž,ž./1[1[3Ð3Ð5E6E8º8º:/;/=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚noqrqrsäsäuVvVxÈxÈz:{:}¬}¬€‚�‚�„…‡t‡t‰æ‰æ‹XŒXŽÊŽÊ�<‘<“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1ƒ}!1èƒ}!1!¾!1d!¾!1kÅ!1lkÅ!1õÎ!1èõÎ!1+Ò!1´+Ò!1Uç!1@Uç!1ó'"1¼ó'"1=/"1Ä=/"1Ç8"1@Ç8"1ý;"1 ý;"1'Q"1˜'Q"1Å‘"1Å‘"1™"1 ™"1™¢"1˜ ™¢"1Ï¥"1d Ï¥"1Ó"1ð Ó"1‹Ú"1Ì ‹Ú"1ëþ"1H ëþ"1-,#1h-,#1C:#1DC:#1ÿn#1Ðÿn#1›�#1ô›�#1™˜#1 ™˜#1Ã$1¬Ã$1›ò$1p›ò$1ÕÞ%1tÕÞ%1Oì%1TOì%1ù€&1àù€&1ÉR'1$ÉR'1qe'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Þª b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 \I‰Ò bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Þª b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 \I‰Ò bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}Þª  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 \I‰Ò  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p�1g‡ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1I‰NÒ[›ä$  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dIÉŽËK’ˆÛ[EŤ$mí6¶ÿÈH‘ZÚ#£ìl µ 5 ~ þ G Ç  � ß  Ù  b ¢ ë +t´ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Ö ÖMMÄÄ);*;4²4²>)?)IrIrST^’^’his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÖÖMMÄÄ ; ; ² ²))  ŽŽ||óój j"á"á$X%X'Ï'Ï)F*F,½,½.4/41«1«3"4"6™6™89;‡;‡=þ=þ?u@uBìBìDcEcGÚGÚIQJQLÆLÆN;O;Q°Q°S%T%VnVnXãXãZX[X]Í]Í_B`BbãbãdXeXgÍgÍiBjBl·l·n6o6q«q«s t v•v•x y {{}ô}ôi€i‚Þ‚Þ„S…S‡ȇȉ=Š=Œ²Œ²Ž'�'‘œ‘œ“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1~!1è~!1±¾!1d±¾!1kÅ!1lkÅ!1õÎ!1èõÎ!1+Ò!1´+Ò!1åç!1@åç!1ƒ("1¼ƒ("1=/"1Ä=/"1Ç8"1@Ç8"1ý;"1 ý;"1·Q"1˜·Q"1U’"1U’"1™"1 ™"1™¢"1˜ ™¢"1Ï¥"1d Ï¥"1Ó"1ð Ó"1aÚ"1Ì aÚ"1Áþ"1H Áþ"1,#1h,#1C:#1DC:#1ÿn#1Ðÿn#1›�#1ô›�#1™˜#1 ™˜#1½$1¬½$1•ò$1p•ò$1ÏÞ%1tÏÞ%1Iì%1TIì%1ó€&1àó€&1ÃR'1$ÃR'1ke'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—Ëm9 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I‰›ÛÒ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—Ëm9 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I‰›ÛÒ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—Ëm9  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I‰›ÛÒ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¹ÙŒ¬@` bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1I‰ LÒ[›ä$  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dIÉ ŒÇG’‚Û[=½¤$mí6¶ÿÈH‘ZÚ#£ìl µ 5 ~ þ G Ç À  � Ù  b ¢ ë +t´ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff à àZZÐÐ)F*F4¼4¼>2?2I¦I¦ST^�^�hØhØrxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffààZZÔÔ N N È ÈBB¼¼66°°**¤¤ ˜ ˜"#%ˆ%ˆ'þ'þ)t*t,ê,ê.`/`1Ö1Ö3L4L6Â6Â8898;®;®=$>$@š@šBCE†E†GüGüIrJrLæLæNZOZQÎQÎSBTBV¶V¶X*Y*[ž[ž]^`†`†búbúdnengâgâiVjVlÊlÊn@o@q´q´s(t(vœvœxy{X{X}Ì}Ì@€@‚´‚´„(…(‡ȇȉ<Š<Œ°Œ°Ž$�$‘˜‘˜“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1ƒ}!1èƒ}!1!¾!1d!¾!1kÅ!1lkÅ!1õÎ!1èõÎ!1+Ò!1´+Ò!1Uç!1@Uç!1ó'"1¼ó'"1=/"1Ä=/"1Ç8"1@Ç8"1ý;"1 ý;"1'Q"1˜'Q"1Å‘"1Å‘"1™"1 ™"1™¢"1˜ ™¢"1Ï¥"1d Ï¥"1Ó"1ð Ó"1mÚ"1Ì mÚ"1Íþ"1H Íþ"1,#1h,#1C:#1DC:#1ÿn#1Ðÿn#1›�#1ô›�#1™˜#1 ™˜#1«$1¬«$1ƒò$1pƒò$1½Þ%1t½Þ%17ì%1T7ì%1�€&1à�€&1]R'1$]R'1e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ܨ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [I‰Ò bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ܨ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [I‰Ò bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ܨ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [I‰Ò  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‘±DdLl bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1I‰‹ËÒ[›ä$  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dIÉ’Û[¤$‚mí¿?6¶|üÿ9¹ÈH‘ZÚ#£ìl µ 5 ~ þ G Ç  � \ œ Ù  b ¢ ë +t´ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ö ÖLLÁÁ)6*64«4«> ? IfIfST^�^�hôhôrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††üürrèè^ ^"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I$J$L—L—N O Q}Q}SðSðU6V6X©X©Z[]�]�_`b¢b¢degˆgˆiûiûknlnnénép\q\sÏsÏuBvBxµxµz({(}›}›€‚�‚�„ô„ô†g‡g‰Ú‰Ú‹MŒMŽÀŽÀ�3‘3“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1~!1è~!1·¾!1d·¾!1kÅ!1lkÅ!1õÎ!1èõÎ!1+Ò!1´+Ò!1åç!1@åç!1ƒ("1¼ƒ("1=/"1Ä=/"1Ç8"1@Ç8"1ý;"1 ý;"1·Q"1˜·Q"1U’"1U’"1™"1 ™"1™¢"1˜ ™¢"1Ï¥"1d Ï¥"1Ó"1ð Ó"17Ú"1Ì 7Ú"1—þ"1H —þ"1Ù+#1hÙ+#1C:#1DC:#1ÿn#1Ðÿn#1›�#1ô›�#1™˜#1 ™˜#1¥$1¬¥$1}ò$1p}ò$1·Þ%1t·Þ%11ì%1T1ì%1�€&1à�€&1QR'1$QR'1ùd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1™Íq= b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I‰šÚÒ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1™Íq= b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I‰šÚÒ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1™Íq=  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I‰šÚÒ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¹Ù„¤8 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1I‰‡ÇÒ[›ä$  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dIÉ’Û[¤$øxmí³36¶oïÿ*ªÈH‘ZÚ#£ìl µ 5 ~ þ G Ç  � ; { Ù  b ¢ ë +t´ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜTTÈÈ)<*<4°4°>$?$I–I–ST^~^~hÃhÃrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœŒŒ | |"ô"ô$h%h'Ü'Ü)P*P,Ä,Ä.8/81¬1¬3 4 6”6”89;|;|=ð=ð?d@dBØBØDLELGÀGÀI4J4L¦L¦NOQŠQŠSüSüUnVnXàXàZR[R]Ä]Ä_6`6b¨b¨degŒgŒiþiþkplpnænæpXqXsÊsÊuŒ>ްް�"‘"“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1‰}!1Љ}!1'¾!1L'¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1[ç!1([ç!1ù'"1¤ù'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1-Q"1€-Q"1Ë‘"1üË‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1=Ú"1´ =Ú"1�þ"10 �þ"1ß+#1Pß+#1]2#1,]2#1õŠ#1˜õŠ#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ß« b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TI‰Ó bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ß« b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TI‰Ó bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~ß«  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TI‰Ó  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 t”¹Ù8 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€¼<xø4´Ð   €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ý ÝSSÉÉ)?*?4µ4µ>)?)I�I�ST^…^…hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÝÝWWÑÑ K K Å Å??¹¹33­­##™™ … …"û"û$q%q'ç'ç)]*],Ó,Ó.I/I1¿1¿35456«6«8!9!;—;—= > @�@�BõBõDiEiGÝGÝIQJQLÅLÅN9O9Q­Q­S!T!V•V•X Y [}[}]ñ]ñ_e`ebÙbÙdMeMgÁgÁi5j5l©l©noq‘q‘stvyvyxíxíza{a}Õ}ÕI€I‚½‚½„1…1‡¥‡¥‰ŠŒ‹Œ‹ŽþŽþ�q‘q“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1 ~!1Ð ~!1«¾!1L«¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1ßç!1(ßç!1}("1¤}("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1±Q"1€±Q"1O’"1üO’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1]2#1,]2#1ûŠ#1˜ûŠ#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1{€&1Ì{€&1KR'1KR'1ód'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀb. b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹ØŸß bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀb. b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹ØŸß bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀb.  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹ØŸß  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‘±SsFf bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1C€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ƒÀ@}ý:ºÓ   €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84­4­> ? I“I“ST^y^yhìhìr_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?L@LB¿B¿D2E2G¥G¥IJL‹L‹NþNþPqQqSäSäUWVWXÊXÊZ=[=]°]°_#`#b–b–d e g|g|iïiïkblbnÕnÕpHqHs»s»u.v.x¡x¡z{}‡}‡úú�m‚m„à„à†S‡S‰ʼnŋ7Œ7ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1‰}!1Љ}!1'¾!1L'¾!1qÅ!1TqÅ!1ûÎ!1ÐûÎ!11Ò!1œ1Ò!1Uç!1(Uç!1ó'"1¤ó'"1C/"1¬C/"1Í8"1(Í8"1<"1ô<"1'Q"1€'Q"1Å‘"1üÅ‘"1™"1 ™"1Ÿ¢"1€ Ÿ¢"1Õ¥"1L Õ¥"1Ó"1Ø Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1Ù4#1,Ù4#1s‹#1˜s‹#1Ÿ$1$Ÿ$1wò$1èwò$1±Þ%1ì±Þ%1]€&1Ì]€&1-R'1-R'1Õd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ý© b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UI‰Ò bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ý© b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UI‰Ò bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ý©  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UI‰Ò  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ”´¹ÙNn bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1|¼€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ðp¬,hè$¤L Œ  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛOOÃÃ)7*74«4«>?I�I�ST^s^shåhårWsW} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››ƒƒ÷÷k k"ß"ß$S%S'Ç'Ç);*;,¯,¯.#/#1—1—3 4 668ó8ó:g;g=Û=Û?M@MB¿B¿D1E1G£G£IJL‡L‡NùNùPkQkSÝSÝUOVOXÁXÁZ3[3]¥]¥_`b‰b‰dûdûfmgmißißkQlQnÃnÃp5q5s§s§uvx‹x‹zýzý|o}oáá�S‚S„ńņ7‡7‰¨‰¨‹ŒŽŠŽŠ�û�û’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1~!1Ð~!1±¾!1L±¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1ßç!1(ßç!1}("1¤}("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1±Q"1€±Q"1O’"1üO’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1Ù4#1,Ù4#1y‹#1˜y‹#1“$1$“$1kò$1èkò$1¥Þ%1ì¥Þ%1€&1Ì€&1åQ'1åQ'1�d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂf2 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 LŒÚ à bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂf2 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 LŒÚ à bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂf2  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 LŒÚ à  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 …¥,L: bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1€À€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1÷w´4qñ.®P �  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙLL¿¿)2*24¥4¥>?I‡I‡SøSø]i^ihÚhÚrKsK} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰üüooââU U"È"È$;%;'®'®)!*!,”,”./1z1z3í3í5`6`8Ó8Ó:F;F=¹=¹?*@*B›B›D E G}G}IîIîK_L_NÐNÐPAQAS²S²U#V#X”X”Z[]v]v_ç_çaXbXdÉdÉf:g:i«i«kln�n�pþpþrosouàuàwQxQzÂzÂ|3}3¤¤�‚„†„††÷†÷ˆg‰g‹׋×�GŽG�·�·’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1�}!1Ð�}!1-¾!1L-¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1aç!1(aç!1ÿ'"1¤ÿ'"1U/"1¬U/"1ß8"1(ß8"1<"1ô<"13Q"1€3Q"1Ñ‘"1üÑ‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1CÚ"1´ CÚ"1£þ"10 £þ"1å+#1På+#1]2#1,]2#1©‹#1˜©‹#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1)�&1Ì)�&1ùR'1ùR'1¡e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kଠb-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 QK‹à  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kଠb-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 QK‹à  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kଠ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 QK‹à   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w—¹Ù> bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÇGŽUÕ€ À  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ü ÜUUÎÎ)G*G4À4À>9?9I°I°S%T%^˜^˜h i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÜÜUUÎÎ G G À À99²²++¤¤–– ˆ ˆ"#%z%z'ó'ó)l*l,å,å.^/^1×1×3P4P6É6É8B9B;»;»=4>4@­@­B&C&EŸEŸGHJ‘J‘LMOOQöQöSmTmVäVäX[Y[[Ò[Ò]I^I`À`Àb5c5eªeªghj”j”l m o|o|qïqïsbtbvÕvÕxHyH{»{»}.~.€¡€¡‚ƒ…‡…‡‡ú‡ú‰lŠlŒތގP�P‘‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1~!1Ð~!1¥¾!1L¥¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Óç!1(Óç!1q("1¤q("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1¥Q"1€¥Q"1C’"1üC’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1�Ú"1´ �Ú"1ýþ"10 ýþ"1?,#1P?,#1]2#1,]2#1m‹#1˜m‹#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µW# b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V–�Ðî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µW# b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V–�Ðî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µW#  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V–�Ðî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 –¶Llp� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÅEŠ OÏ@ €  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞSSÈÈ)=*=4²4²>(?(IœIœST^„^„høhørnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžˆˆýýr r"ç"ç$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡އމŠŒxŒxŽíŽí�b‘b“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1�}!1Ð�}!1-¾!1L-¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1[ç!1([ç!1ù'"1¤ù'"1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1-Q"1€-Q"1Ë‘"1üË‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1å4#1,å4#13Œ#1˜3Œ#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1·€&1Ì·€&1‡R'1‡R'1/e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÞª b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 PJŠÞ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÞª b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 PJŠÞ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1KÞª  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 PJŠÞ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 —·¹ÙPp bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1‡Ç€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1œãcª*qñ G  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>-?-I¢I¢ST^†^†h÷h÷risi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbåbådXeXgËgËi>j>l±l±n"o"q“q“stvuvuxæxæzW{W}È}È9€9‚ª‚ª„…‡�‡�‰þ‰þ‹oŒoŽàŽà�Q‘Q“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1~!1Ð~!1¥¾!1L¥¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1Ùç!1(Ùç!1w("1¤w("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1«Q"1€«Q"1I’"1üI’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1å4#1,å4#1ñ‹#1˜ñ‹#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·[' b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—�Ïð0 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·[' b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—�Ïð0 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·['  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—�Ïð0  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 †¦Qq< bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1…Å€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1”ÙYžcãÅ   €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94®4®>#?#I–I–ST^z^zhìhìr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?O@OBÄBÄD9E9G®G®I#J#L–L–N O Q|Q|SïSïUbVbXÕXÕZH[H]»]»_.`.b b deg„g„iöiökhlhnÚnÚpLqLs¾s¾u0v0x¢x¢z{}†}†øø�j‚j„܄܆P‡P‰ÉË6Œ6ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1•}!1Е}!13¾!1L3¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1aç!1(aç!1ÿ'"1¤ÿ'"1y/"1¬y/"19"1(9"19<"1ô9<"13Q"1€3Q"1Ñ‘"1üÑ‘"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#1c2#1,c2#1]Œ#1˜]Œ#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1;�&1Ì;�&1 S'1 S'1³e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€á­ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Ia¡@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€á­ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Ia¡@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€á­  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Ia¡@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¹Ùzš$D bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÒR¤$vö0 p  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSËË)C*C4»4»>3?3I©I©ST^•^•h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$k%k'ã'ã)[*[,Ó,Ó.K/K1Ã1Ã3;4;6³6³8+9+;£;£=>@“@“B C EƒEƒGûGûIsJsLéLéN_O_QÕQÕSKTKVÁVÁX7Y7[­[­]#^#`™`™bce…e…gûgûiqjqlçlçn]o]qÓqÓsItIv¿v¿x5y5{«{«}!~!€—€—‚ ƒ …ƒ…ƒ‡ù‡ù‰oŠoŒåŒåŽ[�[‘ёѓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1~!1Ð~!1Ÿ¾!1LŸ¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Íç!1(Íç!1k("1¤k("1[/"1¬[/"1å8"1(å8"1<"1ô<"1ŸQ"1€ŸQ"1=’"1ü=’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1©Ú"1´ ©Ú"1 ÿ"10  ÿ"1K,#1PK,#1c2#1,c2#1!Œ#1˜!Œ#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ó€&1Ìó€&1ÃR'1ÃR'1ke'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªL b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V–”Ôâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªL b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V–”Ôâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªL  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V–”Ôâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ›»Rry™ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÐP  pðð 0  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ý ÝTTËË)B*B4¹4¹>2?2I©I©S T ^’^’hisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÝÝTTËË B B ¹ ¹00§§••  ƒƒúúq q"è"è$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIZJZLÑLÑNHOHQ¿Q¿S6T6V­V­X$Y$[›[›]^`‰`‰bcewewgîgîiejelÜlÜnNoNqÀqÀs2t2v¤v¤xy{ˆ{ˆ}ú}úl€l‚Þ‚Þ„P…P‡‡‰4Š4Œ¦Œ¦Ž�‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1�}!1Ð�}!1-¾!1L-¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1aç!1(aç!1ÿ'"1¤ÿ'"1y/"1¬y/"19"1(9"19<"1ô9<"13Q"1€3Q"1Ñ‘"1üÑ‘"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1÷4#1,÷4#1óŒ#1˜óŒ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€ß« b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  J` þ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€ß« b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  J` þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€ß«  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  J` þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¹ÙšºRr bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1’Ò€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1HÈšìl¾>  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙOOÅÅ);*;4±4±>'?'I›I›ST^ƒ^ƒh÷h÷rksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰ÿÿuuëëa a"×"×$M%M'Ã'Ã)9*9,¯,¯.%/%1›1›346‡6‡8ý8ý:s;s=é=é?_@_BÕBÕDKEKGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡‹‡‹‰ÿ‰ÿ‹sŒsŽçŽç�[‘[“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1~!1Ð~!1Ÿ¾!1LŸ¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Óç!1(Óç!1q("1¤q("1[/"1¬[/"1å8"1(å8"1<"1ô<"1¥Q"1€¥Q"1C’"1üC’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1�Ú"1´ �Ú"1ýþ"10 ýþ"1?,#1P?,#1ñ4#1,ñ4#1±Œ#1˜±Œ#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬P b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—•Õä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬P b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—•Õä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬P  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—•Õä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‡§>lŒ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Ѐ� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@À�à`°0€ À  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ù ÙNNÃÃ)8*84­4­>$?$I™I™ST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÙÙNNÃà 8 8 ­ ­""——  ��öökkààU U"Ê"Ê$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1›}!1Л}!19¾!1L9¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1gç!1(gç!1("1¤("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"19Q"1€9Q"1ב"1üב"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1[Ú"1´ [Ú"1»þ"10 »þ"1ý+#1Pý+#1i2#1,i2#1�#1˜�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1�&1Ì�&1ÕR'1ÕR'1}e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�â® b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 FI‰Ò bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�â® b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 FI‰Ò bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�â®  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 FI‰Ò  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¹Ù}�*J bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€Ý]º:—à  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>-?-I¢I¢ST^Œ^Œhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbçbçd\e\gÑgÑiFjFl»l»n0o0q¥q¥stv�v�xy{y{y}î}îc€c‚؂؄M…M‡‡‰7Š7Œ¬Œ¬Ž!�!‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1õ}!1Ðõ}!1“¾!1L“¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Çç!1(Çç!1e("1¤e("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1©Ú"1´ ©Ú"1 ÿ"10  ÿ"1K,#1PK,#1i2#1,i2#1ÕŒ#1˜ÕŒ#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1�&1Ì�&1ÕR'1ÕR'1}e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I‰Ò‹Ë bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I‰Ò‹Ë bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸA   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I‰Ò‹Ë  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  Àr’Xx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€Û[¶6‘  à  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ä ä^^ÔÔ)J*J4À4À>6?6IªIªST^’^’hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjää^^ØØ R R Ì ÌFFÀÀ::´´..¨¨"" œ œ"#%Œ%Œ'(*x*x,î,î.d/d1Ú1Ú3P4P6Æ6Æ8<9<;²;²=(>(@ž@žBCEŠEŠGHJvJvLêLêN^O^QÒQÒSFTFVºVºX.Y.[¢[¢]^`Š`ŠbþbþdrergægæiZjZlÎlÎnBoBq¶q¶s*t*vžvžxy{†{†}ú}ún€n‚â‚â„V…V‡ʇʉ>Š>Œ²Œ²Ž&�&‘š‘š“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1•}!1Е}!13¾!1L3¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1gç!1(gç!1("1¤("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"19Q"1€9Q"1ב"1üב"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#15#1,5#1³�#1˜³�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ଠb-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 EI‰Ò bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ଠb-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 EI‰Ò bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ଠ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 EI‰Ò  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¹Ù�½Tt bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Ý€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1tôQÑ.® ‹}½ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÅÅ):*:4¯4¯>$?$I—I—S T ^}^}hðhðrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=ß=ß?T@TBÉBÉD>E>G³G³I(J(L›L›NOQ�Q�SôSôUgVgXÚXÚZM[M]À]À_3`3b¦b¦degŒgŒiÿiÿkrlrnånåpXqXsËsËu>v>x±x±z${$}—}— € ‚}‚}„ð„ð†c‡c‰Ö‰Ö‹IŒI޼޼�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1û}!1Ðû}!1™¾!1L™¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1Íç!1(Íç!1k("1¤k("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1ŸQ"1€ŸQ"1=’"1ü=’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#15#1,5#1q�#1˜q�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡E b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I‰ÒŒÌ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡E b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I‰ÒŒÌ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡E  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I‰ÒŒÌ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ˆ¨ @m� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1›Û€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1lìGÇ"¢ý};{ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àXXÌÌ)@*@4´4´>(?(IšIšS T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  �� € €"ø"ø$l%l'à'à)T*T,È,È.Œ>ްް�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1›}!1Л}!19¾!1L9¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1mç!1(mç!1 ("1¤ ("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1?Q"1€?Q"1Ý‘"1üÝ‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1i2#1,i2#1Y�#1˜Y�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ç€&1Ìç€&1·R'1·R'1_e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚㯠b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @JŠÓ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚㯠b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @JŠÓ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚㯠 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @JŠÓ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 € 0PZz bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€áaÂB£# ` €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á áWWÍÍ)C*C4¹4¹>-?-I¡I¡ST^‰^‰hýhýrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±''�� ‰ ‰"ÿ"ÿ$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@…@…BùBùDmEmGáGáIUJULÉLÉN=O=Q±Q±S%T%V™V™X Y [�[�]õ]õ_i`ibÝbÝdQeQgÅgÅi9j9l­l­n!o!q•q•s t v}v}xñxñze{e}Ù}ÙM€M‚Á‚Á„5…5‡©‡©‰ŠŒ�Œ�Ž�‘u‘u“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1ï}!1Ðï}!1�¾!1L�¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1Áç!1(Áç!1_("1¤_("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1i2#1,i2#1‰�#1˜‰�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”6 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I‰Ò�Ï bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”6 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I‰Ò�Ï bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”6  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I‰Ò�Ï  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¹Ù¥Å^~ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€æfÌL²2P� €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝRRÇÇ)<*<4±4±>$?$I—I—S T ^}^}hðhðrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��wwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?P@PBÃBÃD6E6G©G©IJL�L�NOQuQuSèSèU[V[XÎXÎZA[A]´]´_'`'bšbšd e g€g€ióiókflfnÙnÙpLqLs¿s¿u2v2x¥x¥z{}‹}‹þþ�q‚q„ä„ä†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1›}!1Л}!19¾!1L9¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1mç!1(mç!1 ("1¤ ("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1?Q"1€?Q"1Ý‘"1üÝ‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#15#1,5#1û�#1˜û�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1o€&1Ìo€&1?R'1?R'1çd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚á­ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @JŠÓ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚á­ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @JŠÓ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚á­  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @JŠÓ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1  ÀKkVv bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¡á€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1„eåFÆ'§Á €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßSSÇÇ);*;4¯4¯>!?!I“I“ST^w^whéhér[s[} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ‡‡ûûo o"ã"ã$W%W'Ë'Ë)?*?,³,³.'/'1›1›346ƒ6ƒ8÷8÷:k;k=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b�b�dÿdÿfqgqiãiãkUlUnÇnÇp9q9s«s«uvx�x�z{}s}såå�W‚W„ɄɆ;‡;‰¬‰¬‹ŒŽŽŽŽ�ÿ�ÿ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1õ}!1Ðõ}!1“¾!1L“¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1Çç!1(Çç!1e("1¤e("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#15#1,5#1+Ž#1˜+Ž#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1-€&1Ì-€&1ýQ'1ýQ'1¥d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–: b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I‰ÒŽÎ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–: b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I‰ÒŽÎ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–:  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I‰ÒŽÎ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¹Ù‰©!A bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¦æ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1˜~þdäJÊö6 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝPPÃÃ)6*64©4©>?I‹I‹SüSü]m^mhÞhÞrOsO} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��ssææY Y"Ì"Ì$?%?'²'²)%*%,˜,˜. / 1~1~3ñ3ñ5d6d8×8×:J;J=½=½?.@.BŸBŸDEG�G�IòIòKcLcNÔNÔPEQES¶S¶U'V'X˜X˜Z [ ]z]z_ë_ëa\b\dÍdÍf>g>i¯i¯k l n‘n‘pqssssuäuäwUxUzÆzÆ|7}7¨¨�‚„ЄІû†ûˆk‰k‹Û‹Û�KŽK�»�»’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1¡}!1С}!1?¾!1L?¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1sç!1(sç!1("1¤("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1EQ"1€EQ"1ã‘"1üã‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1o2#1,o2#1 Ž#1˜ Ž#1á$1$á$1¹ò$1è¹ò$1óÞ%1ìóÞ%1;�&1Ì;�&1 S'1 S'1³e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒä° b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @JŠÓ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒä° b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @JŠÓ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒä°  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @JŠÓ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ƒ£Ss6V bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ìlØXÄDÐ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg à àYYÒÒ)K*K4Ä4Ä>=?=I´I´S)T)^œ^œhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggààYYÒÒ K K Ä Ä==¶¶//¨¨!!šš Œ Œ"#%~%~'÷'÷)p*p,é,é.b/b1Û1Û3T4T6Í6Í8F9F;¿;¿=8>8@±@±B*C*E£E£GHJ•J•L M OƒOƒQúQúSqTqVèVèX_Y_[Ö[Ö]M^M`Ä`Äb9c9e®e®g#h#j˜j˜l m o€o€qóqósftfvÙvÙxLyL{¿{¿}2~2€¥€¥‚ƒ…‹…‹‡þ‡þ‰pŠpŒâŒâŽT�T‘ƑƓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1é}!1Ðé}!1‡¾!1L‡¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1»ç!1(»ç!1Y("1¤Y("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1�Q"1€�Q"1+’"1ü+’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1o2#1,o2#1=Ž#1˜=Ž#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1Û€&1ÌÛ€&1«R'1«R'1Se'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰+÷ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I‰Ò–Ö bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰+÷ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I‰Ò–Ö bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰+÷  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I‰Ò–Ö  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¹Ù¢ÂTt bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ñqâbÓS@ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âWWÌÌ)A*A4¶4¶>,?,I I ST^ˆ^ˆhühürrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢ŒŒ v v"ë"ë$`%`'Õ'Õ)J*J,¿,¿.4/41©1©346“6“89;};}=ò=ò?g@gBÜBÜDQEQGÆGÆI1?1I¦I¦ST^Š^Šhûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIYJYLÎLÎNCOCQ¸Q¸S-T-V¢V¢XY[Œ[Œ]^`v`vbébéd\e\gÏgÏiBjBlµlµn&o&q—q—stvyvyxêxêz[{[}Ì}Ì=€=‚®‚®„…‡‘‡‘‰ŠŒsŒsŽäŽä�U‘U“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1ï}!1Ðï}!1�¾!1L�¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1»ç!1(»ç!1Y("1¤Y("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1�Q"1€�Q"1+’"1ü+’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#1!5#1,!5#1ëŽ#1˜ëŽ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹/û b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 I‰Ò—× bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹/û b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 I‰Ò—× bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹/û  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 I‰Ò—×  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¹Ù‚¢3 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1±ñ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÄDµ5¦&—±ñ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞSSÈÈ)=*=4²4²>'?'IšIšS T ^~^~hðhðrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽxxííb b"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¤b¤degˆgˆiúiúklllnÞnÞpPqPsÂsÂu4v4x¦x¦z{}Š}Šüü�n‚n„à„à†T‡T‰ljNj:Œ:Ž­Ž­� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1¿}!1�¿}!1]¾!1]¾!1³Â!1$³Â!1‹ç!1�‹ç!1)("1)("1…,"1$…,"1]Q"1�]Q"1û‘"1û‘"1W–"1$W–"1…Ú"1�…Ú"1åþ"1üåþ"1/0#1 /0#1û~#1| û~#1ù$1Ü ù$1Ñò$1  Ñò$1 ß%1¤  ß%1_�&1„ _�&1/S'1È/S'1×e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Qa•öÂ÷à b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Qa•öÂ÷à b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Qa•öÂ÷à b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Aa  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx ð ðhhàà)X*X4Ð4Ð>F?FI¼I¼S2T2^¨^¨his”s”} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðxxððhhàà X X Ð ÐHHÀÀ88°°((   � �"#%€%€'ø'ø)p*p,è,è.`/`1Ø1Ø3P4P6È6È8@9@;¸;¸=0>0@¨@¨B C E˜E˜GHJ†J†LüLüNrOrQèQèS^T^VÔVÔXJYJ[À[À]6^6`¬`¬b"c"e˜e˜ghj„j„lúlúnpopqæqæs\t\vÒvÒxHyH{¾{¾}4~4€ª€ª‚ ƒ …–…–‡ ˆ ЂЂŒøŒøŽn�n‘ä‘ä“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1ã}!1�ã}!1�¾!1 �¾!1YÅ!1YÅ!1µç!1€µç!1S("1üS("1+/"1+/"1‡Q"1p‡Q"1%’"1ì%’"1ý˜"1ôý˜"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1/0#1ì/0#1û~#1L û~#1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ë b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¥å€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ë b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¥å€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}ë b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¥å €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¹Ù  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss ê êaaØØ)O*O4Æ4Æ>=?=I´I´S+T+^�^�his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssêêaaØØ O O Æ Æ==´´++¢¢�� ~ ~"õ"õ$l%l'ã'ã)Z*Z,Ñ,Ñ.H/H1¿1¿36466­6­8$9$;›;›=>@‰@‰BCEwEwGîGîIeJeLÜLÜNSOSQÊQÊSATAV¸V¸X/Y/[¦[¦]^`”`”b c e‚e‚gùgùipjplçlçnYoYqËqËs=t=v¯v¯x!y!{“{“}~€w€w‚é‚é„[…[‡͇͉?Š?Œ±Œ±Ž#�#‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1¹}!1�¹}!1W¾!1W¾!1³Â!1$³Â!1‹ç!1�‹ç!1)("1)("1…,"1$…,"1]Q"1�]Q"1û‘"1û‘"1W–"1$W–"1‹Ú"1�‹Ú"1ëþ"1üëþ"1/0#1 /0#1û~#1| û~#1á$1Ü á$1¹ò$1  ¹ò$1óÞ%1¤ óÞ%1ç€&1„ ç€&1·R'1È·R'1_e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R_“ò¾úÆ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R_“ò¾úÆ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R_“ò¾úÆ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Hh  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ì ìbbØØ)N*N4Ä4Ä>8?8I¬I¬S T ^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvììbbØØ N N Ä Ä::°°&&œœˆˆþþt t"ê"ê$`%`'Ö'Ö)L*L,Â,Â.8/81®1®3$4$6š6š89;†;†=ü=ü?r@rBèBèD^E^GÔGÔIHJHL¼L¼N0O0Q¤Q¤STVŒVŒXY[t[t]è]è_\`\bÐbÐdDeDg¸g¸i,j,l l noqˆqˆsüsüupvpxäxäzX{X}Ì}Ì@€@‚´‚´„(…(‡œ‡œ‰ŠŒ„Œ„ŽøŽø�l‘l“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1é}!1�é}!1‡¾!1 ‡¾!1YÅ!1YÅ!1µç!1€µç!1S("1üS("1+/"1+/"1‡Q"1p‡Q"1%’"1ì%’"1ý˜"1ôý˜"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1/0#1ì/0#1û~#1L û~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1·€&1T ·€&1‡R'1˜‡R'1/e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}!í b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¤ä€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}!í b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¤ä€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1I}!í b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¤ä €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¹Ù  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss è è]]ÒÒ)G*G4¼4¼>1?1I¦I¦ST^�^�hÿhÿrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssèè]]ÒÒ G G ¼ ¼11¦¦��zzïïd d"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I)J)LžLžNOQˆQˆSýSýUrVrXçXçZ\[\]Ñ]Ñ_F`Fb»b»d0e0g¥g¥ijl�l�noqsqssåsåuWvWxÉxÉz;{;}­}­€‚‘‚‘„…‡u‡u‰ç‰ç‹YŒYŽËŽË�=‘=“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1³}!1�³}!1Q¾!1 Q¾!1³Â!1³Â!1…ç!1€…ç!1#("1ü#("1…,"1…,"1WQ"1pWQ"1õ‘"1ìõ‘"1W–"1ôW–"1yÚ"1`yÚ"1Ùþ"1ÌÙþ"1/0#1ì/0#1û~#1L û~#1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1�&1T �&1íR'1˜íR'1•e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ó¿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ó¿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ó¿ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1:Z  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ë ëbbÙÙ)P*P4Ç4Ç>>µµ,,££‘‘  "ö"ö$m%m'ä'ä)[*[,Ò,Ò.I/I1À1À37476®6®8%9%;œ;œ=>@Š@ŠBCExExGïGïIdJdLÙLÙNNONQÃQÃS8T8V­V­X"Y"[—[—] ^ `�`�böbödkekgàgàiUjUlÊlÊn?o?q´q´s)t)vžvžxy{ˆ{ˆ}ý}ýr€r‚ç‚ç„\…\‡чщFŠFŒ»Œ»Ž0�0‘¥‘¥“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1é}!1�é}!1‡¾!1 ‡¾!1GÅ!1GÅ!1»ç!1€»ç!1Y("1üY("1/"1/"1�Q"1p�Q"1+’"1ì+’"1ë˜"1ôë˜"1ùØ"1`ùØ"1Yý"1ÌYý"1/0#1ì/0#1û~#1L û~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1�&1T �&1çR'1˜çR'1�e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†(ô b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1–Ö€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†(ô b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1–Ö€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†(ô b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1–Ö €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¹Ù  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss í íggÝÝ)S*S4É4É>=?=I±I±S%T%^™^™h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssííggáá [ [ Õ ÕOOÉÉCC½½77±±++ ¥ ¥"#%•%•' ( *�*�,÷,÷.m/m1ã1ã3Y4Y6Ï6Ï8E9E;»;»=1>1@§@§BCE“E“G H J}J}LñLñNeOeQÙQÙSMTMVÁVÁX5Y5[©[©]^`‘`‘bceyeygígíiajalÕlÕnIoIq½q½s1t1v¥v¥xy{�{�}~€u€u‚é‚é„]…]‡чщEŠEŒ¹Œ¹Ž-�-‘¡‘¡“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1³}!1�³}!1Q¾!1 Q¾!1³Â!1³Â!1ç!1€ç!1("1ü("1…,"1…,"1QQ"1pQQ"1ï‘"1ìï‘"1W–"1ôW–"1‹Ú"1`‹Ú"1ëþ"1Ìëþ"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1·€&1T ·€&1‡R'1˜‡R'1/e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ï» b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ï» b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ï» b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Kk  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss é é__ÔÔ)I*I4¾4¾>1?1I¤I¤ST^Š^Šhýhýrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðsséé__ÕÕ K K Á Á77­­##™™……ûûq q"ç"ç$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI5J5L¨L¨NOQŽQŽSTVtVtXçXçZZ[Z]Í]Í_@`@b³b³d&e&g™g™i j llnònòpeqesØsØuKvKx¾x¾z1{1}¤}¤€‚ЂЄý„ý†p‡p‰ã‰ã‹VŒVŽÉŽÉ�<‘<“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1ï}!1�ï}!1�¾!1 �¾!1GÅ!1GÅ!1»ç!1€»ç!1Y("1üY("1/"1/"1�Q"1p�Q"1+’"1ì+’"1ë˜"1ôë˜"1ùØ"1`ùØ"1Yý"1ÌYý"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†*ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1—×€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†*ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1—×€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†*ö b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1—× €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¹Ù  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss ë ëcc××)K*K4¿4¿>1?1I£I£ST^‡^‡hùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssëëccÛÛ S S Ë ËCC»»33««##›› ‹ ‹"#%w%w'ë'ë)_*_,Ó,Ó.G/G1»1»3/4/6£6£89;‹;‹=ÿ=ÿ?s@sBçBçD[E[GÏGÏIAJAL³L³N%O%Q—Q—S T V{V{XíXíZ_[_]Ñ]Ñ_C`Cbµbµd'e'g™g™i j l}l}nïnïpaqasÓsÓuEvEx·x·z){)}›}› € ‚‚„ñ„ñ†c‡c‰Õ‰Õ‹GŒG޹޹�+‘+“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1­}!1�­}!1K¾!1 K¾!1¹Â!1¹Â!1ç!1€ç!1("1ü("1‹,"1‹,"1QQ"1pQQ"1ï‘"1ìï‘"1]–"1ô]–"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1ó€&1T ó€&1ÃR'1˜ÃR'1ke'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ð¼ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ð¼ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ð¼ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¹Ù  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt î îddÚÚ)P*P4Ä4Ä>8?8I¬I¬S T ^”^”hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttîîhhââ \ \ Ö ÖPPÊÊDD¾¾44ªª  – –" # %‚%‚'ø'ø)n*n,ä,ä.Z/Z1Ð1Ð3F4F6¼6¼8292;¨;¨=>@�@�BCExExGìGìI`J`LÔLÔNHOHQ¼Q¼S0T0V¤V¤XY[Œ[Œ]^`t`tbèbèd\e\gÐgÐiDjDl¸l¸n,o,q q stvˆvˆxüxüzp{p}ä}äX€X‚̂̄@…@‡²‡²‰%Š%Œ˜Œ˜Ž � ‘~‘~“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1ï}!1�ï}!1�¾!1 �¾!1MÅ!1MÅ!1Áç!1€Áç!1_("1ü_("1/"1/"1“Q"1p“Q"11’"1ì1’"1ñ˜"1ôñ˜"1—Ú"1`—Ú"1÷þ"1Ì÷þ"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�1ý b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ÿß€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�1ý b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ÿß€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�1ý b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ÿß €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1\|  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll â âWWÌÌ)A*A4´4´>'?'IšIšS T ^€^€hóhórdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllââXXÎÎ D D º º00¦¦’’||ññf f"Û"Û$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=à=à?S@SBÆBÆD9E9G¬G¬IJL’L’NOQxQxSëSëU^V^XÑXÑZD[D]·]·_*`*b�b�degƒgƒiöiökilinÜnÜpOqOsÂsÂu5v5x¨x¨z{}Ž}Ž€‚t‚t„ç„ç†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1­}!1�­}!1K¾!1 K¾!1¹Â!1¹Â!1yç!1€yç!1("1ü("1‹,"1‹,"1KQ"1pKQ"1é‘"1ìé‘"1]–"1ô]–"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1{€&1T {€&1KR'1˜KR'1ód'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ì¸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ì¸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ì¸ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¹Ù  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ê ê^^ÒÒ)F*F4¸4¸>*?*IœIœST^€^€hòhòrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrêêbbÚÚ R R Ê ÊBBºº22ªª’’ z z"î"î$b%b'Ö'Ö)J*J,¾,¾.2/21¦1¦346Ž6Ž89;v;v=è=è?Z@ZBÌBÌD>E>G°G°I"J"L”L”NOQxQxSêSêU\V\XÎXÎZ@[@]²]²_$`$b–b–degzgziìiìk^l^nÐnÐpBqBs´s´u&v&x˜x˜z { }|}|îî�`‚`„Ò„Ò†B‡B‰³‰³‹$Œ$ޕޕ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1ï}!1�ï}!1�¾!1 �¾!1MÅ!1MÅ!1Áç!1€Áç!1_("1ü_("1/"1/"1“Q"1p“Q"11’"1ì1’"1ñ˜"1ôñ˜"1£Ú"1`£Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%13€&1T 3€&1R'1˜R'1«d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�3ÿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1žÞ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�3ÿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1žÞ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�3ÿ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1žÞ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1g‡  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn ä äWWÊÊ)=*=4®4®>?I�I�ST^r^rhãhãrRsR} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnääZZÐÐ F F ¼ ¼22¨¨””zzíí` `"Ó"Ó$F%F'¹'¹),*,,Ÿ,Ÿ./1…1…3ø3ø5k6k8Þ8Þ:Q;Q=Â=Â?3@3B¤B¤DEG†G†I÷I÷KhLhNÙNÙPJQJS»S»U,V,X�X�Z[]]_ð_ðaabadÒdÒfCgCi´i´k%l%n–n–pqsxsxuéuéwZxZzËzË|<}<­­�‚„�„�†þ†þˆn‰n‹Þ‹Þ�NŽN�¾�¾’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1§}!1�§}!1E¾!1 E¾!1¹Â!1¹Â!1yç!1€yç!1("1ü("1‹,"1‹,"1KQ"1pKQ"1é‘"1ìé‘"1]–"1ô]–"1óØ"1`óØ"1Sý"1ÌSý"1#0#1ì#0#1ï~#1L ï~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1;�&1T ;�&1 S'1˜ S'1³e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒí¹ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1R€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒí¹ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1R€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒí¹ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1R €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¹Ù  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp é ébbÛÛ)T*T4Í4Í>D?DI»I»S0T0^£^£his‡s‡} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppéébbÛÛ T T Í ÍFF¿¿88±±**££ • •"#%‡%‡'(*y*y,ò,ò.k/k1ä1ä3]4]6Ö6Ö8O9O;È;È=A>A@º@ºB3C3E¬E¬G%H%JœJœLMOŠOŠQRTxTxVïVïXfYf[Ý[Ý]T^T`Ë`Ëb@c@eµeµg*h*jŸjŸlmo‡o‡qúqúsmtmvàvàxSyS{Æ{Æ}9~9€¬€¬‚ƒ…’…’‡ˆŠuŠuŒçŒçŽY�Y‘ˑ˓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1õ}!1�õ}!1“¾!1 “¾!1;Å!1;Å!1Çç!1€Çç!1e("1üe("1 /"1 /"1™Q"1p™Q"17’"1ì7’"1ߘ"1ôߘ"1�Ú"1`�Ú"1ýþ"1Ìýþ"1#0#1ì#0#1ï~#1L ï~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Õ€&1T Õ€&1¥R'1˜¥R'1Me'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜: b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‘Ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜: b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‘Ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜: b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‘Ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1cƒ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk ã ãXXÍÍ)B*B4·4·>+?+IŸIŸST^‡^‡hûhûroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkãã[[ÓÓ K K à Ã;;³³++££�� w w"ì"ì$a%a'Ö'Ö)K*K,À,À.5/51ª1ª346”6”8 9 ;~;~=ó=ó?h@hBÝBÝDRERGÇGÇI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÃbÃd7e7g«g«ijl“l“noq{q{sïsïucvcx×x×zK{K}¿}¿3€3‚§‚§„…‡�‡�‰ŠŒyŒyŽîŽî�c‘c“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¡}!1�¡}!1?¾!1 ?¾!1¿Â!1¿Â!1sç!1€sç!1("1ü("1‘,"1‘,"1EQ"1pEQ"1ã‘"1ìã‘"1c–"1ôc–"1óØ"1`óØ"1Sý"1ÌSý"1#0#1ì#0#1ï~#1L ï~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠéµ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1S€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠéµ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1S€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠéµ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1S €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¹Ù  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn å å\\ÓÓ)J*J4Á4Á>6?6I«I«ST^�^�hispsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnåå\\ÓÓ J J Á Á88¯¯&&��‹‹ y y"ð"ð$g%g'Þ'Þ)U*U,Ì,Ì.C/C1º1º31416¨6¨89;–;–= > @„@„BûBûDrErGéGéI^J^LÓLÓNHOHQ½Q½S2T2V§V§XY[‘[‘]^`{`{bîbîdaeagÔgÔiGjGlºlºn+o+qœqœs t v~v~xïxïz`{`}Ñ}ÑB€B‚³‚³„$…$‡”‡”‰ŠŒvŒvŽçŽç�X‘X“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1õ}!1�õ}!1“¾!1 “¾!1;Å!1;Å!1Çç!1€Çç!1e("1üe("1 /"1 /"1™Q"1p™Q"17’"1ì7’"1ߘ"1ôߘ"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜< b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1’Ò€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜< b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1’Ò€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜< b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1’Ò €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1r’  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk á áVVËË)@*@4µ4µ>(?(I›I›S T ^^hñhñrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkááWWÍÍ C C ¹ ¹//¥¥‘‘{{ððe e"Ú"Ú$O%O'Ä'Ä)9*9,®,®.#/#1˜1˜3 4 6‚6‚8÷8÷:l;l=á=á?V@VBËBËD@E@GµGµI(J(L›L›NOQ�Q�SôSôUgVgXÚXÚZM[M]À]À_3`3b¥b¥deg‰g‰iûiûkmlmnßnßpQqQsÃsÃu5v5x§x§z{}‹}‹ýý�o‚o„á„á†S‡S‰ƉƋ9Œ9ެެ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¡}!1�¡}!1?¾!1 ?¾!1¹Â!1¹Â!1sç!1€sç!1("1ü("1‹,"1‹,"1EQ"1pEQ"1ã‘"1ìã‘"1]–"1ô]–"1íØ"1`íØ"1Mý"1ÌMý"1#0#1ì#0#1ï~#1L ï~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1G�&1T G�&1S'1˜S'1¿e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ê¶ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ê¶ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ê¶ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¹Ù  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ä ä\\ÔÔ)L*L4Ä4Ä>:?:I°I°S&T&^œ^œhisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllää\\ÔÔ L L Ä Ä<<´´,,¤¤””  „ „"ü"ü$t%t'ì'ì)d*d,Ü,Ü.T/T1Ì1Ì3D4D6¼6¼8494;¬;¬=$>$@œ@œBCEŒEŒGHJzJzLðLðNfOfQÜQÜSRTRVÈVÈX>Y>[´[´]*^*` ` bceŒeŒghjxjxlîlîndodqÚqÚsPtPvÆvÆx1?1I¨I¨ST^‘^‘hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIYJYLÐLÐNGOGQ¾Q¾S5T5V¬V¬X#Y#[š[š]^`ˆ`ˆbÿbÿdvevgígíidjdlÛlÛnMoMq¿q¿s1t1v£v£xy{‡{‡}ù}ùk€k‚݂݄O…O‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1›}!1�›}!19¾!1 9¾!1³Â!1³Â!1mç!1€mç!1 ("1ü ("1…,"1…,"1?Q"1p?Q"1Ý‘"1ìÝ‘"1W–"1ôW–"1íØ"1`íØ"1Mý"1ÌMý"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1Ï€&1T Ï€&1ŸR'1˜ŸR'1Ge'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡æ² b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡æ² b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡æ² b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¹Ù  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj à àVVÌÌ)B*B4¸4¸>,?,I I ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjààVVÌÌ B B ¸ ¸..¤¤��||òòh h"Þ"Þ$T%T'Ê'Ê)@*@,¶,¶.,/,1¢1¢346Ž6Ž89;z;z=ð=ð?f@fBÜBÜDRERGÈGÈI%?%IšIšST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÜÜQQÆÆ ; ; ° °%%šš„„ùùnnããX X"Í"Í$B%B'·'·),*,,¡,¡./1‹1‹346u6u8ê8ê:_;_=Ô=Ô?I@IB¾B¾D3E3G¨G¨IJL’L’NOQ|Q|SñSñUfVfXÛXÛZP[P]Å]Å_:`:b¯b¯d$e$g™g™ijlƒlƒnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1›}!1�›}!19¾!1 9¾!1³Â!1³Â!1gç!1€gç!1("1ü("1…,"1…,"19Q"1p9Q"1ב"1ìב"1W–"1ôW–"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ç³ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1O€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ç³ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1O€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ç³ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1O €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ºÚ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>0?0I¥I¥ST^�^�hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubêbêd_e_gÔgÔiIjIl¾l¾n3o3q¨q¨stv’v’xy{|{|}ñ}ñf€f‚Û‚Û„P…P‡Ňʼn:Š:Œ¯Œ¯Ž$�$‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1~!1�~!1Ÿ¾!1 Ÿ¾!1/Å!1/Å!1Íç!1€Íç!1k("1ük("1/"1/"1ŸQ"1pŸQ"1=’"1ì=’"1Ó˜"1ôÓ˜"1‘Ú"1`‘Ú"1ñþ"1Ìñþ"10#1ì0#1é~#1L é~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªL b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ŽÎ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªL b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ŽÎ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªL b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ŽÎ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1a�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á á[[ÑÑ)G*G4½4½>1?1I¥I¥ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±++¥¥ ™ ™"#%‰%‰'ÿ'ÿ)u*u,ë,ë.a/a1×1×3M4M6Ã6Ã8999;¯;¯=%>%@›@›BCE‡E‡GýGýIqJqLåLåNYOYQÍQÍSATAVµVµX)Y)[�[�]^`…`…bùbùdmemgágáiUjUlÉlÉn=o=q±q±s%t%v™v™x y {�{�}õ}õi€i‚݂݄Q…Q‡Ňʼn9Š9Œ­Œ­Ž!�!‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1•}!1�•}!13¾!1 3¾!1³Â!1³Â!1gç!1€gç!1("1ü("1…,"1…,"19Q"1p9Q"1ב"1ìב"1W–"1ôW–"1çØ"1`çØ"1Gý"1ÌGý"10#1ì0#1é~#1L é~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„㯠b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„㯠b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„㯠b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¹Ù  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝSSÈÈ)=*=4²4²>%?%I˜I˜S T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��yyïïe e"Û"Û$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnænæpYqYsÌsÌu?v?x²x²z%{%}˜}˜ € ‚~‚~„ñ„ñ†d‡d‰׉׋JŒJ޽޽�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1~!1�~!1Ÿ¾!1 Ÿ¾!1/Å!1/Å!1Óç!1€Óç!1q("1üq("1/"1/"1¥Q"1p¥Q"1C’"1ìC’"1Ó˜"1ôÓ˜"1µÚ"1`µÚ"1ÿ"1Ìÿ"10#1ì0#1é~#1L é~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªN b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ï€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªN b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ï€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªN b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ï €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1†¦  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßWWËË)?*?4³4³>%?%I—I—S T ^{^{híhír_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ��  "÷"÷$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnãnãpUqUsÇsÇu9v9x«x«z{}�}�€‚s‚s„å„å†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1•}!1�•}!13¾!1 3¾!1­Â!1­Â!1aç!1€aç!1ÿ'"1üÿ'"1,"1,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1Q–"1ôQ–"1IÚ"1`IÚ"1©þ"1Ì©þ"10#1ì0#1ã~#1L ã~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒä° b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒä° b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒä° b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1;  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh â âXXÎÎ)D*D4¸4¸>,?,I I ST^Š^Šhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²((žž Š Š"#%v%v'ì'ì)b*b,Ø,Ø.N/N1Ä1Ä3:4:6°6°8&9&;œ;œ=>@„@„BøBøDlElGàGàITJTLÈLÈN?IŽIŽS T ^~^~hñhñrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††ûûppååZ Z"Ï"Ï$D%D'¹'¹).*.,£,£./1�1�346w6w8ì8ì:a;a=Ô=Ô?G@GBºBºD-E-G G IJL†L†NùNùPlQlSßSßURVRXÅXÅZ8[8]«]«_`b›b›deg�g�iôiôkglgnÚnÚpMqMsÀsÀu3v3x¦x¦z{}Œ}Œÿÿ�r‚r„å„å†V‡V‰ȉȋ:Œ:ެެ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1�}!1��}!1-¾!1 -¾!1­Â!1­Â!1aç!1€aç!1ÿ'"1üÿ'"1,"1,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1Q–"1ôQ–"1gÚ"1`gÚ"1Çþ"1ÌÇþ"10#1ì0#1ã~#1L ã~#1¥$1¬ ¥$1}ò$1p }ò$1·Þ%1t ·Þ%1c€&1T c€&13R'1˜3R'1Ûd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ଠb-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ଠb-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ଠb-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations18X  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞRRÆÆ):*:4¬4¬>?I�I�ST^x^xhêhêrZsZ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žž††úún n"â"â$V%V'Ê'Ê)>*>,²,².&/&1š1š346‚6‚8ö8ö:j;j=Ü=Ü?N@NBÀBÀD2E2G¤G¤IJLˆLˆNúNúPlQlSÞSÞUPVPXÂXÂZ4[4]¦]¦_`bŽbŽdegrgriäiäkVlVnÈnÈp:q:s¬s¬uvx�x�z{}t}tææ�X‚X„ʄʆ:‡:‰«‰«‹ŒŽ�Ž��þ�þ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1~!1�~!1¥¾!1 ¥¾!15Å!15Å!1Óç!1€Óç!1q("1üq("1/"1/"1¥Q"1p¥Q"1C’"1ìC’"1Ù˜"1ôÙ˜"1çØ"1`çØ"1Gý"1ÌGý"10#1ì0#1ã~#1L ã~#1™$1¬ ™$1qò$1p qò$1«Þ%1t «Þ%1!€&1T !€&1ñQ'1˜ñQ'1™d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³W# b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1˜Ø€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³W# b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1˜Ø€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³W# b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1˜Ø €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¹Ù  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØKK¾¾)1*14¢4¢>?I„I„SýSý]n^nhßhßrNsN} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆûûnnááT T"Ç"Ç$:%:'­'­) * ,“,“./1y1y3ì3ì5_6_8Ò8Ò:E;E=¶=¶?'@'B˜B˜D E GzGzIëIëK\L\NÍNÍP>Q>S¯S¯U V X‘X‘Z[]s]s_ä_äa]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1‰}!1�‰}!1'¾!1 '¾!1§Â!1§Â!1[ç!1€[ç!1ù'"1üù'"1y,"1y,"1-Q"1p-Q"1Ë‘"1ìË‘"1K–"1ôK–"1½Ø"1`½Ø"1ý"1Ìý"10#1ì0#1ã~#1L ã~#1Õ$1¬ Õ$1ýð$1p ýð$1ÛÀ%1t ÛÀ%1/�&1T /�&1%Q'1˜%Q'1Íc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€á­ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€á­ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€á­ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¹µ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ý ÝVVÑÑ)J*J4Ã4Ã>:?:I±I±S&T&^¥^¥his‰s‰} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÝÝVVÏÏ H H Á Á::³³,,¥¥—— ‰ ‰"#%}%}'ö'ö)o*o,è,è.a/a1Ú1Ú3S4S6Ì6Ì8E9E;¾;¾=7>7@°@°B)C)E¢E¢GHJ’J’L M O€O€Q÷Q÷SnTnVåVåX\Y\[Ó[Ó]J^J`Á`Áb6c6e«e«g h j•j•l m o‰o‰qüqüsotovâvâxUyU{È{È};~;€®€®‚!ƒ!…”…”‡ˆŠwŠwŒéŒéŽ[�[‘͓͑ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1~!1�~!1¥¾!1 ¥¾!1#Å!1#Å!1Ùç!1€Ùç!1w("1üw("1õ."1õ."1«Q"1p«Q"1I’"1ìI’"1ǘ"1ôǘ"1Ú"1`Ú"1yþ"1Ôyþ"10#1ô0#1ã~#1T ã~#1Ã$1´ Ã$1ëð$1x ëð$1ÉÀ%1| ÉÀ%1Ï€&1\ Ï€&1ÅP'1 ÅP'1mc'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼^* b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‹Ë€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼^* b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‹Ë€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼^* b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‹Ë €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1¥í\  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ × ×LLËË)@*@4µ4µ>)?)I�I�ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__××OOÇÇ ? ? · ·//§§——  ��öök k"à"à$_%_'Ô'Ô)I*I,¾,¾.3/31¨1¨346’6’89;|;|=ñ=ñ?f@fBÛBÛDPEPGÅGÅI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÍbÍdAeAgµgµi)j)l�l�noq…q…sùsùumvmxáxázU{U}É}É=€=‚±‚±„%…%‡™‡™‰ŠŒƒŒƒŽøŽø�m‘m“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1‰}!1�‰}!1'¾!1 '¾!1§Â!1§Â!1Uç!1€Uç!1ó'"1üó'"1y,"1y,"1'Q"1p'Q"1Å‘"1ìÅ‘"1K–"1ôK–"1½Ø"1`½Ø"1ý"1Ìý"10#1ì0#1ã~#1L ã~#1½$1¬ ½$1åð$1p åð$1ÃÀ%1t ÃÀ%1½€&1T ½€&1³P'1˜³P'1[c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ý© b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ý© b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1J~Ý© b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¹µ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPËË)B*B4¹4¹>.?.I£I£ST^“^“histst} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbæbædYeYgÌgÌi?j?l²l²n/o/q q stv‚v‚xóxózd{d}Õ}ÕF€F‚·‚·„(…(‡˜‡˜‰ Š ŒzŒzŽëŽë�\‘\“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1 ~!1� ~!1«¾!1 «¾!1#Å!1#Å!1Ùç!1€Ùç!1w("1üw("1õ."1õ."1«Q"1p«Q"1I’"1ìI’"1ǘ"1ôǘ"1%Ú"1`%Ú"1…þ"1Ô…þ"10#1ô0#1ã~#1T ã~#1·$1´ ·$1ßð$1x ßð$1½À%1| ½À%1‡€&1\ ‡€&1}P'1 }P'1%c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼`, b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ŒÌ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼`, b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ŒÌ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼`, b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ŒÌ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1¦îk  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Õ ÕJJÇÇ)<*<4±4±>$?$I—I—ST^‡^‡hùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……úúooääY Y"Î"Î$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD4?4I«I«S"T"^ ^ his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÕÕLLÃà : : ƒ ƒúúqqèè__{{òòi i"à"à$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãI\J\LÓLÓNJOJQÁQÁS8T8V¯V¯X&Y&[�[�]^`‹`‹bceyeygðgðigjglÞlÞn\o\qÎqÎs@t@v²v²x$y${–{–}~€z€z‚ì‚ì„^…^‡ЇЉBŠBŒ´Œ´Ž&�&‘˜‘˜“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1~!1è~!1½¾!1d½¾!1kÅ!1lkÅ!1õÎ!1èõÎ!1+Ò!1´+Ò!1ñç!1@ñç!1�("1¼�("1=/"1Ä=/"1Ç8"1@Ç8"1ý;"1 ý;"1ÃQ"1˜ÃQ"1a’"1a’"1™"1 ™"1™¢"1˜ ™¢"1Ï¥"1d Ï¥"1Ó"1ð Ó"1OÚ"1Ì OÚ"1¯þ"1H ¯þ"1ñ+#1hñ+#1C:#1DC:#1ÿn#1Ðÿn#1¿�#1ô¿�#1½˜#1 ½˜#1Ï$1¬Ï$1§ò$1p§ò$1áÞ%1táÞ%1[ì%1T[ì%1;�&1à;�&1 S'1$ S'1³e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¤Ø|H b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹‰ÉÖ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¤Ø|H b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹‰ÉÖ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¤Ø|H  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹‰ÉÖ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‘±»Û2R bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1K‹ÖNa¡ì,  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dKË–áa¬,w÷B �ØXˆ£#ÇGnî†9¹EÅ „ Ï O š  e å 0 ° û ; ÿ ? † Æ QœÜ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSÍÍ)*4½4½>5?5I«I«S!T!^£^£his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$m%m'å'å)]*],Õ,Õ.M/M1—1—346‡6‡8ÿ8ÿ:w;w=>@•@•B C E…E…GýGýIuJuLëLëNaOaQ×Q×SMTMVÃVÃX9Y9[¯[¯]%^%`›`›bce‡e‡gýgýisjslélénkokqáqásWtWvÍvÍxCyC{¹{¹}/~/€¥€¥‚ƒ…‘…‘‡ˆŠ}Š}ŒóŒóŽi�i‘ߑߓ  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„+i!1M}!1M}!1ë½!1|ë½!1UÃ!1„UÃ!1‹Æ!1ð‹Æ!1Ð!1|Ð!1KÓ!1HKÓ!1ç!1Ôç!1½'"1P½'"1'-"1X'-"1]0"1Ä]0"1ç9"1Pç9"1="1="1ñP"1¨ñP"1�‘"1$ �‘"1ù–"1, ù–"1/š"1˜ /š"1¹£"1$ ¹£"1ï¦"1ð ï¦"11Ô"1| 1Ô"1=Ú"1X=Ú"1�þ"1Ô�þ"1ß+#1ôß+#1£:#1У:#1_o#1\_o#1Ÿ�#1€Ÿ�#1�š#1¬�š#1·$18·$1�ò$1ü�ò$1ÉÞ%1ÉÞ%1Cì%1àCì%1™€&1l™€&1iR'1°iR'1e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O°| b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1»û bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O°| b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1»û  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O°|  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1»û bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ^~>ì  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1d¤HªêŠÊ LŒ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1ddäHÈ,¬�ôtØX¼<  ‚däF Æ  � À @ ( ¨  ÿ  Š > ¾ ì l Î N ° 0z º ”Ô8xÚ|¼ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ ¤ ¤HHÅÅ):*:4¯4¯>&?&I›I›ST^Ž^Žhisrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÓÓHH½½ 2 2 x xííbb××LLððeeÚÚO O"Ä"Ä$A%A'¶'¶)+*+, , ./1Š1Š3ÿ3ÿ5t6t8é8é:^;^=Ó=Ó?H@HB½B½D2E2G§G§IJL“L“NOQ}Q}SòSòUgVgXÜXÜZQ[Q]Æ]Æ_;`;b°b°d%e%gšgšijl„l„noqtqtsæsæuXvXxÊxÊz<{<}®}® € ‚’‚’„…‡v‡v‰è‰è‹ZŒZŽÌŽÌ�>‘>“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1%~!1è%~!1þ!1dþ!1kÅ!1lkÅ!1õÎ!1èõÎ!1+Ò!1´+Ò!1ñç!1@ñç!1�("1¼�("1=/"1Ä=/"1Ç8"1@Ç8"1ý;"1 ý;"1ÃQ"1˜ÃQ"1a’"1a’"1™"1 ™"1™¢"1˜ ™¢"1Ï¥"1d Ï¥"1Ó"1ð Ó"1Ú"1Ì Ú"1yþ"1H yþ"1»+#1h»+#1C:#1DC:#1ÿn#1Ðÿn#1¿�#1ô¿�#1½˜#1 ½˜#1½$1¬½$1•ò$1p•ò$1ÏÞ%1tÏÞ%1Iì%1TIì%1À&1àÀ&1“R'1$“R'1;e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¦Ú€L b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹ŠÊÖ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¦Ú€L b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹ŠÊÖ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¦Ú€L  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹ŠÊÖ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 »Ûe…ú bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1K‹Öa¡�Í ì,  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dKË–áa¬,w÷B �ØX£#nî9¹ „  … Ï O Ä D š  ƒ  e å B  0 ° û ; † Æ ~ ¾ QœÜ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×MMÇÇ)*4³4³>)?)I�I�ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$O%O'Å'Å);*;,±,±.'/'1n1n3ä3ä5Z6Z8Ð8Ð:F;F=ë=ë?a@aB×B×DMEMGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noq…q…sùsùumvmxáxázU{U}É}É=€=‚±‚±„%…%‡™‡™‰ Š Œ�Œ�ŽõŽõ�i‘i“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1‰}!1è‰}!1'¾!1d'¾!1qÅ!1lqÅ!1ûÎ!1èûÎ!11Ò!1´1Ò!1[ç!1@[ç!1ù'"1¼ù'"1C/"1ÄC/"1Í8"1@Í8"1<"1 <"1-Q"1˜-Q"1Ë‘"1Ë‘"1™"1 ™"1Ÿ¢"1˜ Ÿ¢"1Õ¥"1d Õ¥"1Ó"1ð Ó"1‘Ú"1Ì ‘Ú"1ñþ"1H ñþ"13,#1h3,#1I:#1DI:#1o#1Ðo#1Å�#1ôÅ�#1Ø#1 Ø#1Ã$1¬Ã$1›ò$1p›ò$1ÕÞ%1tÕÞ%1Oì%1TOì%1ù€&1àù€&1ÉR'1$ÉR'1qe'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ä° b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 XK‹Ö bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ä° b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 XK‹Ö bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�ä°  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 XK‹Ö  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‰©"BnŽ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1K‹ÖFa¡ì,  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dKË–áa¬,w÷B �ØXôt£#±1nîoï9¹,¬ „ Ï O š  e å 0 ° û ; ×  † Æ QœÜ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ü ÜVVÌÌ)B*B4¸4¸>.?.ItItST^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÜÜVVÐÐ J J Ä Ä>>¸¸22¬¬&&   ” ”"#%„%„'ú'ú)p*p,æ,æ.\/\1Ò1Ò3H4H6¾6¾8494;ª;ª= > @–@–B C E‚E‚GøGøInJnLâLâNVOVQÊQÊS>T>V„V„XøXøZl[l]à]à_T`TböbödjejgÞgÞiRjRlÆlÆnDoDq¸q¸s,t,v v xy{ˆ{ˆ}ü}üp€p‚ä‚ä„X…X‡̇̉@Š@Œ´Œ´Ž(�(‘œ‘œ“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1~!1è~!1·¾!1d·¾!1qÅ!1lqÅ!1ûÎ!1èûÎ!11Ò!1´1Ò!1ëç!1@ëç!1‰("1¼‰("1C/"1ÄC/"1Í8"1@Í8"1<"1 <"1½Q"1˜½Q"1[’"1[’"1™"1 ™"1Ÿ¢"1˜ Ÿ¢"1Õ¥"1d Õ¥"1Ó"1ð Ó"1OÚ"1Ì OÚ"1¯þ"1H ¯þ"1ñ+#1hñ+#1I:#1DI:#1o#1Ðo#1Å�#1ôÅ�#1Ø#1 Ø#1½$1¬½$1•ò$1p•ò$1ÏÞ%1tÏÞ%1Iì%1TIì%1ó€&1àó€&1ÃR'1$ÃR'1ke'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1™Íq= b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹�ÝÖ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1™Íq= b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹�ÝÖ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1™Íq=  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹�ÝÖ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 »Û�°0P bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1K‹Öý=a¡ì,  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dKË–áa¬,w÷B �ØXëk£#¨(nîeå9¹"¢ „ Ï O š  e å 0 ° û ; Í † Æ QœÜ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>-?-I¢I¢ST^Ž^ŽhÕhÕrxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbçbçd\e\gÑgÑiFjFl»l»n2o2q§q§stv‘v‘xy{M{M}Â}Â7€7‚¬‚¬„!…!‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1‰}!1è‰}!1'¾!1d'¾!1qÅ!1lqÅ!1ûÎ!1èûÎ!11Ò!1´1Ò!1Uç!1@Uç!1ó'"1¼ó'"1C/"1ÄC/"1Í8"1@Í8"1<"1 <"1'Q"1˜'Q"1Å‘"1Å‘"1™"1 ™"1Ÿ¢"1˜ Ÿ¢"1Õ¥"1d Õ¥"1Ó"1ð Ó"1gÚ"1Ì gÚ"1Çþ"1H Çþ"1 ,#1h ,#1I:#1DI:#1o#1Ðo#1Ë�#1ôË�#1ɘ#1 ɘ#1«$1¬«$1ƒò$1pƒò$1½Þ%1t½Þ%17ì%1T7ì%1�€&1à�€&1QR'1$QR'1ùd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kଠb-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 WK‹Ö bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kଠb-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 WK‹Ö bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Kଠ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 WK‹Ö  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 cƒöBb bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1K‹Öa¡ƒà ì,  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dKË–áa¬,w÷BÂ�ÚZ¥%pð<¼ ‡ éi Ò R ¦ & �  c ã   h è 3 ³ þ > ‰ É T ” TŸß bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚRRÆÆ):*:4®4®>"?"IeIeST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""ššŠŠ z z"ò"ò$f%f'Ú'Ú)N*N,Â,Â.6/61ª1ª346’6’89;z;z=î=î?b@bBÖBÖDJEJG¾G¾I2J2L¤L¤NOQˆQˆSúSúU=V=X¯X¯Z![!]“]“_`b¦b¦degŠgŠiüiüknlnnènèpZqZsÌsÌu>v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1~!1è~!1½¾!1d½¾!1qÅ!1lqÅ!1ûÎ!1èûÎ!11Ò!1´1Ò!1ëç!1@ëç!1‰("1¼‰("1C/"1ÄC/"1Í8"1@Í8"1<"1 <"1½Q"1˜½Q"1[’"1[’"1™"1 ™"1Ÿ¢"1˜ Ÿ¢"1Õ¥"1d Õ¥"1Ó"1ð Ó"1%Ú"1Ì %Ú"1…þ"1H …þ"1Ç+#1hÇ+#1I:#1DI:#1o#1Ðo#1Å�#1ôÅ�#1Ø#1 Ø#1«$1¬«$1ƒò$1pƒò$1½Þ%1t½Þ%17ì%1T7ì%1�€&1à�€&1]R'1$]R'1e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1›ÏuA b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹œÜÖ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1›ÏuA b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹œÜÖ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1›ÏuA  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹œÜÖ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 »ÛjŠ$ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1K‹Öa¡zº ì,  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dKË–áa¬,w÷B �ØX£#nî9¹ „ ß_ Ï O œ  š  Y Ù  – e å 0 ° û ; † Æ J Š QœÜ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÃÃ)8*84­4­>"?"I•I•ST^^hÃhÃrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD?I�I�ST^u^uhèhèr[s[} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……úúooääY Y"Î"Î$C%C'¸'¸)-*-,¢,¢./1Œ1Œ346v6v8ë8ë:`;`=Õ=Õ?H@HB»B»D.E.G¡G¡IJL‡L‡NúNúPmQmSàSàUSVSXÆXÆZ9[9]¬]¬_`b’b’degxgxiëiëk^l^nÑnÑpDqDs·s·u*v*x�x�z{}ƒ}ƒöö�i‚i„܄܆O‡O‰Á‰Á‹3Œ3ޥޥ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1~!1Ð~!1±¾!1L±¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1ßç!1(ßç!1}("1¤}("1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1±Q"1€±Q"1O’"1üO’"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1m7#1,m7#1™Œ#1˜™Œ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Õ€&1ÌÕ€&1¥R'1¥R'1Me'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂf2 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹™ÙÚ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂf2 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹™ÙÚ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂf2  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹™ÙÚ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �¯cƒ.N bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1N€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18¸ý}ÂB‡ n ®  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á áWWÍÍ)C*C4¹4¹>-?-I¡I¡ST^‰^‰hýhýrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±''�� ‰ ‰"ÿ"ÿ$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@…@…BùBùDmEmGáGáIUJULÉLÉN=O=Q±Q±S%T%V™V™X Y [�[�]õ]õ_i`ibÝbÝdQeQgÅgÅi9j9l­l­n!o!q•q•s t v}v}xñxñze{e}Ù}ÙM€M‚Á‚Á„5…5‡©‡©‰ŠŒ�Œ�Ž�‘u‘u“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1�}!1Ð�}!1-¾!1L-¾!1wÅ!1TwÅ!1Ï!1ÐÏ!17Ò!1œ7Ò!1[ç!1([ç!1ù'"1¤ù'"1I/"1¬I/"1Ó8"1(Ó8"1 <"1ô <"1-Q"1€-Q"1Ë‘"1üË‘"1™"1 ™"1¥¢"1€ ¥¢"1Û¥"1L Û¥"1Ó"1Ø Ó"1%Ú"1´ %Ú"1…þ"10 …þ"1Ç+#1PÇ+#1Ñ9#1,Ñ9#1oŒ#1˜oŒ#1“$1$“$1kò$1èkò$1¥Þ%1ì¥Þ%1€&1Ì€&1åQ'1åQ'1�d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€á­ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UK‹Ö bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€á­ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UK‹Ö bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€á­  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UK‹Ö  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 »Ûhˆ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1v¶ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ØX –  T Ô  ’ F †  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×JJ½½)0*04£4£>?I…I…SöSö]g^ghØhØrIsI} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡úúmmààS S"Æ"Æ$9%9'¬'¬)*,’,’./1x1x3ë3ë5^6^8Ñ8Ñ:D;D=·=·?(@(B™B™D E G{G{IìIìK]L]NÎNÎP?Q?S°S°U!V!X’X’Z[]t]t_å_åaVbVdÇdÇf8g8i©i©kln‹n‹püpürmsmuÞuÞwOxOzÀzÀ|1}1¢¢�‚„„„„†õ†õˆe‰e‹Õ‹Õ�EŽE�µ�µ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1~!1Ð~!1±¾!1L±¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1åç!1(åç!1ƒ("1¤ƒ("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1·Q"1€·Q"1U’"1üU’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1ï9#1,ï9#1�#1˜�#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1]€&1Ì]€&1-R'1-R'1Õd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Äj6 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹šÚÛ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Äj6 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹šÚÛ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Äj6  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹šÚÛ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 o�;[. bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1“Ó €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1L Ì  ‘ Ö V ›  ó 3 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝQQÅÅ)9*94­4­>?I‘I‘ST^u^uhçhçrYsY} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��……ùùm m"á"á$U%U'É'É)=*=,±,±.%/%1™1™3 4 6�6�8õ8õ:i;i=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b‹b‹dýdýfogoiáiákSlSnÅnÅp7q7s©s©uvx�x�zÿzÿ|q}qãã�U‚U„DŽdž9‡9‰ª‰ª‹ŒŽŒŽŒ�ý�ý’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1•}!1Е}!13¾!1L3¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1gç!1(gç!1("1¤("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"19Q"1€9Q"1ב"1üב"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#1m7#1,m7#1½Œ#1˜½Œ#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒæ² b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  MM�à  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒæ² b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  MM�à  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒæ²  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  MM�à   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‡§»Û> bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1N€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18¸�ÊJ“ Ž Î  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚOOÄÄ)9*94®4®>$?$I˜I˜S T ^€^€hôhôrjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""šš„„ùùn n"ã"ã$X%X'Í'Í)B*B,·,·.,/,1¡1¡346‹6‹89;u;u=ê=ê?_@_BÔBÔDIEIG¾G¾I4J4L¨L¨NOQ�Q�STVxVxXìXìZ`[`]Ô]Ô_H`Hb¼b¼d0e0g¤g¤ijlŒlŒnoqtqtsèsèu\v\xÐxÐzD{D}¸}¸,€,‚ ‚ „…‡ЇЉÿ‰ÿ‹tŒtŽéŽé�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1 ~!1Ð ~!1«¾!1L«¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Ùç!1(Ùç!1w("1¤w("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1«Q"1€«Q"1I’"1üI’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1¯Ú"1´ ¯Ú"1ÿ"10 ÿ"1Q,#1PQ,#1…7#1,…7#1_�#1˜_�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1)�&1Ì)�&1ùR'1ùR'1¡e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·[' b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T”’Òî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·[' b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T”’Òî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·['  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T”’Òî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ž®,LŸ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1$d€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�`à0° € 4t €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg à àYYÒÒ)K*K4Ä4Ä>=?=I´I´S)T)^œ^œhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggààYYÒÒ K K Ä Ä==¶¶//¨¨!!šš Œ Œ"#%~%~'÷'÷)p*p,é,é.b/b1Û1Û3T4T6Í6Í8F9F;¿;¿=8>8@±@±B*C*E£E£GHJ•J•L M OƒOƒQúQúSqTqVèVèX_Y_[Ö[Ö]M^M`Ä`Äb9c9e®e®g#h#j˜j˜l m o€o€qóqósftfvÙvÙxLyL{¿{¿}2~2€¥€¥‚ƒ…‹…‹‡þ‡þ‰pŠpŒâŒâŽT�T‘ƑƓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1�}!1Ð�}!1-¾!1L-¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1aç!1(aç!1ÿ'"1¤ÿ'"1U/"1¬U/"1ß8"1(ß8"1<"1ô<"13Q"1€3Q"1Ñ‘"1üÑ‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"11Ú"1´ 1Ú"1‘þ"10 ‘þ"1Ó+#1PÓ+#1õ9#1,õ9#1A�#1˜A�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�â® b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  LLŒß bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�â® b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  LLŒß bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�â®  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  LLŒß  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 »Ûm� * bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1—× €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1\ Ü % ¥ î n · 7 W €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØMMÂÂ)7*74¬4¬>!?!I”I”ST^x^xhêhêr^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?M@MBÂBÂD7E7G¬G¬I!J!L”L”NOQzQzSíSíU`V`XÓXÓZF[F]¹]¹_,`,bžbždeg‚g‚iôiôkflfnØnØpJqJs¼s¼u.v.x x z{}„}„öö�h‚h„Ú„Ú†N‡N‰Á‰Á‹4Œ4ާާ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1 ~!1Ð ~!1«¾!1L«¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ßç!1(ßç!1}("1¤}("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1±Q"1€±Q"1O’"1üO’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1:#1,:#1ï�#1˜ï�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1·€&1Ì·€&1‡R'1‡R'1/e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹_+ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•‘Ñð0 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹_+ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•‘Ñð0 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹_+  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•‘Ñð0  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 t”8`€ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1´ô €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ð P   p ð @ À Ä €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>/?/I¤I¤ST^ˆ^ˆhùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbçbçdZeZgÍgÍi@j@l³l³n$o$q•q•stvwvwxèxèzY{Y}Ê}Ê;€;‚¬‚¬„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1›}!1Л}!19¾!1L9¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1mç!1(mç!1 ("1¤ ("1y/"1¬y/"19"1(9"19<"1ô9<"1?Q"1€?Q"1Ý‘"1üÝ‘"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#1‹7#1,‹7#1ƒ�#1˜ƒ�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1í€&1Ìí€&1½R'1½R'1ee'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„ç³ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Ic£@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„ç³ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Ic£@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„ç³  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Ic£@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 »Û†¦< bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1$d€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�dä8¸ Œ T” €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÇÇ)>*>4µ4µ>.?.I¥I¥ST^Ž^Žhisrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝIVJVLÍLÍNDODQ»Q»S2T2V©V©X Y [—[—]^`…`…bübüdsesgêgêiajalØlØnJoJq¼q¼s.t.v v xy{„{„}ö}öh€h‚Ú‚Ú„L…L‡¾‡¾‰0Š0Œ¢Œ¢Ž�‘†‘†“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1~!1Ð~!1Ÿ¾!1LŸ¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Óç!1(Óç!1q("1¤q("1[/"1¬[/"1å8"1(å8"1<"1ô<"1¥Q"1€¥Q"1C’"1üC’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1¯Ú"1´ ¯Ú"1ÿ"10 ÿ"1Q,#1PQ,#1�7#1,�7#1+Ž#1˜+Ž#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1A�&1ÌA�&1S'1S'1¹e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬P b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 T”ŽÎâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬P b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 T”ŽÎâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬P  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 T”ŽÎâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �­*J|œ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1:z€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1èhÃCž y ù ü< €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßWWÏÏ)G*G4¿4¿>7?7I­I­S#T#^™^™his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ��  "÷"÷$o%o'ç'ç)_*_,×,×.O/O1Ç1Ç3?4?6·6·8/9/;§;§=>@—@—BCE‡E‡GÿGÿIwJwLíLíNcOcQÙQÙSOTOVÅVÅX;Y;[±[±]'^'`�`�bce‰e‰gÿgÿiujulëlënaoaq×q×sMtMvÃvÃx9y9{¯{¯}%~%€›€›‚ƒ…‡…‡‡ý‡ý‰sŠsŒéŒéŽ_�_‘Õ‘Õ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1•}!1Е}!13¾!1L3¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1gç!1(gç!1("1¤("1y/"1¬y/"19"1(9"19<"1ô9<"19Q"1€9Q"1ב"1üב"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1CÚ"1´ CÚ"1£þ"10 £þ"1å+#1På+#1:#1,:#1Ž#1˜Ž#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚㯠b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jb¢þ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚㯠b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jb¢þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚㯠 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jb¢þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 »Ûr’4 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¸ø €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1à ` ´ 4 ˆ  \ Ü è( €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb × ×LLÁÁ)6*64«4«>"?"I—I—S T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbb××LLÁÁ 6 6 « «  ••  ôôiiÞÞS S"È"È$=%='²'²)'*',œ,œ./1†1†3û3û5p6p8å8å:Z;Z=Ï=Ï?D@DB¹B¹D.E.G£G£IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1~!1Ð~!1¥¾!1L¥¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Ùç!1(Ùç!1w("1¤w("1[/"1¬[/"1å8"1(å8"1<"1ô<"1«Q"1€«Q"1I’"1üI’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#17:#1,7:#1ÇŽ#1˜ÇŽ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®T  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•�Ïä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®T  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•�Ïä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®T   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•�Ïä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 y™"Bo� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Õ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1T Ô / ¯ Š å e —× €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÇÇ)=*=4³4³>)?)I�I�ST^…^…hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒuŒuŽéŽé�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¡}!1С}!1?¾!1L?¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1mç!1(mç!1 ("1¤ ("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1?Q"1€?Q"1Ý‘"1üÝ‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#1£7#1,£7#1OŽ#1˜OŽ#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1�&1Ì�&1ÕR'1ÕR'1}e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…è´ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 BK‹Ö bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…è´ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 BK‹Ö bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…è´  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 BK‹Ö  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 »Û…¥: bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1:z€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1èhÇG¦& …  Z €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff à àZZÐÐ)F*F4¼4¼>2?2I¦I¦ST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffààZZÔÔ N N È ÈBB¼¼66°°**¤¤ ˜ ˜"#%ˆ%ˆ'þ'þ)t*t,ê,ê.`/`1Ö1Ö3L4L6Â6Â8898;®;®=$>$@š@šBCE†E†GüGüIrJrLæLæNZOZQÎQÎSBTBV¶V¶X*Y*[ž[ž]^`†`†búbúdnengâgâiVjVlÊlÊn>o>q²q²s&t&všvšxy{‚{‚}ö}öj€j‚Þ‚Þ„R…R‡ƇƉ:Š:Œ®Œ®Ž"�"‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1û}!1Ðû}!1™¾!1L™¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1Íç!1(Íç!1k("1¤k("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1ŸQ"1€ŸQ"1=’"1ü=’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1©Ú"1´ ©Ú"1 ÿ"10  ÿ"1K,#1PK,#1»7#1,»7#1÷Ž#1˜÷Ž#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1�&1Ì�&1ÕR'1ÕR'1}e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡E b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹‹Ë× bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡E b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹‹Ë× bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡E  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹‹Ë×  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Œ¬(Hu• bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1P�€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@À&¦ Œ ò r À €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Þ ÞUUÌÌ)C*C4º4º>1?1I¦I¦ST^�^�hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIYJYLÎLÎNCOCQ¸Q¸S-T-V¢V¢XY[Œ[Œ]^`v`vbëbëd`e`gÕgÕiJjJl¿l¿n4o4q©q©stv“v“xy{}{}}ò}òg€g‚܂܄Q…Q‡ƇƉ;Š;Œ°Œ°Ž%�%‘š‘š“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1›}!1Л}!19¾!1L9¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1mç!1(mç!1 ("1¤ ("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1?Q"1€?Q"1Ý‘"1üÝ‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#1C:#1,C:#1ëŽ#1˜ëŽ#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒä° b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 AM�Ø bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒä° b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 AM�Ø bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒä°  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 AM�Ø  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 »Ûw—> bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ù €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1d ä C à " ¢  � ¹ù €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÊÊ)>*>4²4²>&?&I˜I˜S T ^|^|hîhîr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$j%j'Þ'Þ)R*R,Æ,Æ.:/:1®1®3"4"6–6–8 9 ;~;~=ò=ò?f@fBÚBÚDNENGÂGÂI6J6L¨L¨NOQŒQŒSþSþUpVpXâXâZT[T]Æ]Æ_8`8bªbªdegŽgŽijlrlrnänäpVqVsÈsÈu:v:x¬x¬z{}�}�€‚t‚t„æ„æ†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1~!1Ð~!1Ÿ¾!1LŸ¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1Óç!1(Óç!1q("1¤q("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1¥Q"1€¥Q"1C’"1üC’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#1a:#1,a:#1™�#1˜™�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Ÿ€&1ÌŸ€&1oR'1oR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£I b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹‹Ë× bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£I b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹‹Ë× bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£I  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹‹Ë×  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~ž,Lp� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ö6 €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ø X ¾ > ¤ $ Š f¦ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜRRÇÇ)<*<4±4±>&?&I™I™S T ^^hòhòrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒxxîîd d"Ú"Ú$O%O'Ä'Ä)9*9,®,®.#/#1˜1˜3 4 6‚6‚8÷8÷:l;l=á=á?V@VBËBËD@E@GµGµI*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b¨b¨degŽgŽijltltnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1§}!1Ч}!1E¾!1LE¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1sç!1(sç!1("1¤("1[/"1¬[/"1å8"1(å8"1<"1ô<"1EQ"1€EQ"1ã‘"1üã‘"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1¯7#1,¯7#1�Ž#1˜�Ž#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†éµ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @LŒ× bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†éµ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @LŒ× bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†éµ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @LŒ×  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 „¤Ww8 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1B‚€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ˆëkÎN ± 1 b¢ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84­4­> ? I“I“ST^y^yhìhìr_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?L@LB¿B¿D2E2G¥G¥IJL‹L‹NþNþPqQqSäSäUWVWXÊXÊZ=[=]°]°_#`#b–b–d e g|g|iïiïkblbnÕnÕpHqHs»s»u.v.x¡x¡z{}‡}‡úú�m‚m„à„à†S‡S‰ʼnŋ7Œ7ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1õ}!1Ðõ}!1“¾!1L“¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1Çç!1(Çç!1e("1¤e("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1™Q"1€™Q"17’"1ü7’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1[Ú"1´ [Ú"1»þ"10 »þ"1ý+#1Pý+#1Á7#1,Á7#1�#1˜�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1í€&1Ìí€&1½R'1½R'1ee'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–: b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹Ö�Í bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–: b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹Ö�Í bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–:  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹Ö�Í  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 »Û‹«&F bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1T”€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1PÐ8¸    ˆ ä$ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk å å[[ÑÑ)G*G4½4½>1?1I¥I¥ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkåå__ÙÙ S S Í ÍGGÁÁ;;µµ++¡¡ � �"#%y%y'ï'ï)e*e,Û,Û.Q/Q1Ç1Ç3=4=6³6³8)9);Ÿ;Ÿ=>@‰@‰BýBýDqEqGåGåIYJYLÍLÍNAOAQµQµS)T)V�V�XY[…[…]ù]ù_m`mbábádUeUgÉgÉi=j=l±l±n%o%q™q™s t v�v�xõxõzi{i}Ý}ÝQ€Q‚łń9…9‡­‡­‰ Š Œ“Œ“Ž�‘y‘y“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1¡}!1С}!1?¾!1L?¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1sç!1(sç!1("1¤("1[/"1¬[/"1å8"1(å8"1<"1ô<"1EQ"1€EQ"1ã‘"1üã‘"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1O:#1,O:#1?�#1˜?�#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1-€&1Ì-€&1ýQ'1ýQ'1¥d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„å± b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @LŒ× bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„å± b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @LŒ× bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„å±  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @LŒ×  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 |œDd(H bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1å% €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1”  w ÷ Z Ú = ½ E €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛNNÁÁ)4*44§4§>?I‰I‰SúSú]k^khÜhÜrMsM} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹þþqqääW W"Ê"Ê$=%='°'°)#*#,–,–. / 1|1|3ï3ï5b6b8Õ8Õ:H;H=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éaZbZdËdËf  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹Ö�Í bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜>  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹Ö�Í bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜>   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹Ö�Í  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 »Ûƒ£6V bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ü< €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ð p Ø X À @ ¨ (ŒÌ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áUUÉÉ)=*=4±4±>#?#I•I•ST^y^yhëhër]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‰‰ýýq q"å"å$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=á=á?S@SBÅBÅD7E7G©G©IJL�L�NÿNÿPqQqSãSãUUVUXÇXÇZ9[9]«]«_`b�b�degsgsiåiåkWlWnÉnÉp;q;s­s­uvx‘x‘z{}u}uçç�Y‚Y„˄ˆ=‡=‰®‰®‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1§}!1Ч}!1E¾!1LE¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1yç!1(yç!1("1¤("1[/"1¬[/"1å8"1(å8"1<"1ô<"1KQ"1€KQ"1é‘"1üé‘"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1Ç7#1,Ç7#1c�#1˜c�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1Õ€&1ÌÕ€&1¥R'1¥R'1Me'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ê¶ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @LŒ× bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ê¶ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @LŒ× bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ê¶  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @LŒ×  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ƒ£Xx6 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1X˜€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`àNÎ< ¼ * ª (h €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞSSÈÈ)=*=4²4²>(?(IœIœST^„^„høhørnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžˆˆýýr r"ç"ç$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡އމŠŒxŒxŽíŽí�b‘b“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1ï}!1Ðï}!1�¾!1L�¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Áç!1(Áç!1_("1¤_("1[/"1¬[/"1å8"1(å8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1Ù7#1,Ù7#1ç�#1˜ç�#1ç$1$ç$1¿ò$1è¿ò$1ùÞ%1ìùÞ%1A�&1ÌA�&1S'1S'1¹e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹/û b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹Ö”Ô bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹/û b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹Ö”Ô bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹/û  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹Ö”Ô  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 »Û†¦< bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1jª€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¨(› Ž  �  ªê €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk ä ä]]ÖÖ)O*O4È4È>A?AI¸I¸S-T-^ ^ his†s†} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkää]]ÖÖ O O È ÈAAºº33¬¬%%žž � �" # %‚%‚'û'û)t*t,í,í.f/f1ß1ß3X4X6Ñ6Ñ8J9J;Ã;Ã=<><@µ@µB.C.E§E§G H J™J™LMO‡O‡QþQþSuTuVìVìXcYc[Ú[Ú]Q^Q`È`Èb=c=e²e²g'h'jœjœlmo„o„q÷q÷sjtjvÝvÝxPyP{Ã{Ã}6~6€©€©‚ƒ…�…�‡ˆŠtŠtŒæŒæŽX�X‘ʑʓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1§}!1Ч}!1E¾!1LE¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1sç!1(sç!1("1¤("1[/"1¬[/"1å8"1(å8"1<"1ô<"1EQ"1€EQ"1ã‘"1üã‘"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1y:#1,y:#1�#1˜�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…æ² b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @LŒ× bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…æ² b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @LŒ× bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…æ²  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @LŒ×  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �¡Yy2R bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1F €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ˜  † ô t â bÖ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜQQÆÆ);*;4°4°>%?%I˜I˜S T ^|^|hîhîrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒvvëë` `"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1õ}!1Ðõ}!1“¾!1L“¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Áç!1(Áç!1_("1¤_("1[/"1¬[/"1å8"1(å8"1<"1ô<"1“Q"1€“Q"11’"1ü1’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1‹:#1,‹:#1›�#1˜›�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�3ÿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 K‹Ö•Õ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�3ÿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 K‹Ö•Õ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�3ÿ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 K‹Ö•Õ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 »Û„¤8X bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1] €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1t ô g ç Z Ú MÍ]� €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii à àWWÎÎ)E*E4¼4¼>3?3I¨I¨ST^Œ^Œhýhýroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiààWWÎÎ E E ¼ ¼33ªª!!˜˜††ýýt t"ë"ë$b%b'Ù'Ù)P*P,Ç,Ç.>/>1µ1µ3,4,6£6£89;‘;‘=>@@BöBöDmEmGäGäI[J[LÐLÐNEOEQºQºS/T/V¤V¤XY[Ž[Ž]^`x`xbëbëd^e^gÑgÑiDjDl·l·n(o(q™q™s t v{v{xìxìz]{]}Î}Î?€?‚°‚°„!…!‡“‡“‰ŠŒuŒu޿޿�W‘W“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Å}!1�Å}!1c¾!1c¾!1¹Â!1$¹Â!1‘ç!1�‘ç!1/("1/("1‹,"1$‹,"1cQ"1�cQ"1’"1’"1]–"1$]–"1‹Ú"1�‹Ú"1ëþ"1üëþ"150#1 50#1#1| #1í$1Ü í$1Åò$1  Åò$1ÿÞ%1¤ ÿÞ%1�&1„ �&1çR'1ÈçR'1�e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Me™üÈñ½ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Me™üÈñ½ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Me™üÈñ½ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Hh  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww î îeeÜÜ)S*S4Ê4Ê>A?AI¸I¸S/T/^¡^¡his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwîîeeÜÜ S S Ê ÊAA¸¸//¦¦””  ‚ ‚"ù"ù$p%p'ç'ç)^*^,Õ,Õ.L/L1Ã1Ã3:4:6±6±8(9(;Ÿ;Ÿ=>@�@�BCE{E{GòGòIiJiLàLàNWOWQÎQÎSETEV¼V¼X3Y3[ª[ª]!^!`˜`˜bce†e†gýgýitjtlëlën]o]qÏqÏsAtAv³v³x%y%{—{—} ~ €{€{‚í‚í„_…_‡чщCŠCŒµŒµŽ'�'‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1é}!1�é}!1‡¾!1 ‡¾!1YÅ!1YÅ!1»ç!1€»ç!1Y("1üY("1+/"1+/"1�Q"1p�Q"1+’"1ì+’"1ý˜"1ôý˜"1ÿØ"1`ÿØ"1_ý"1Ì_ý"150#1ì50#1#1L #1ù$1¬ ù$1Ñò$1p Ñò$1 ß%1t  ß%1_�&1T _�&1/S'1˜/S'1×e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1K#ï b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1£ã€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1K#ï b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1£ã€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1K#ï b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1£ã €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1»Û  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ì ìddÜÜ)T*T4Ì4Ì>B?BI¸I¸S.T.^¤^¤his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttììddÜÜ T T Ì ÌDD¼¼44¬¬$$œœ Œ Œ"#%|%|'ô'ô)l*l,ä,ä.\/\1Ô1Ô3L4L6Ä6Ä8<9<;´;´=,>,@¤@¤BCE”E”G H J‚J‚LøLøNnOnQäQäSZTZVÐVÐXFYF[¼[¼]2^2`¨`¨bce”e”g h j€j€lölönlolqâqâsXtXvÎvÎxDyD{º{º}0~0€¦€¦‚ƒ…’…’‡ˆŠ~Š~ŒôŒôŽj�j‘à‘à“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1¿}!1�¿}!1]¾!1]¾!1¹Â!1$¹Â!1‹ç!1�‹ç!1)("1)("1‹,"1$‹,"1]Q"1�]Q"1û‘"1û‘"1]–"1$]–"1—Ú"1�—Ú"1÷þ"1ü÷þ"150#1 50#1#1| #1Û$1Ü Û$1³ò$1  ³ò$1íÞ%1¤ íÞ%1½€&1„ ½€&1�R'1È�R'15e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Na•öÂôÀ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Na•öÂôÀ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Na•öÂôÀ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Oo  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu ê ê__ÔÔ)I*I4¾4¾>3?3I¨I¨ST^�^�hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuêê__ÔÔ I I ¾ ¾33¨¨’’||ññf f"Û"Û$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I+J+L L NOQŠQŠSÿSÿUtVtXéXéZ^[^]Ó]Ó_H`Hb½b½d2e2g§g§ijl‘l‘noququsçsçuYvYxËxËz={=}¯}¯!€!‚“‚“„…‡w‡w‰é‰é‹[Œ[ŽÍŽÍ�?‘?“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1ï}!1�ï}!1�¾!1 �¾!1YÅ!1YÅ!1»ç!1€»ç!1Y("1üY("1+/"1+/"1�Q"1p�Q"1+’"1ì+’"1ý˜"1ôý˜"1ÿØ"1`ÿØ"1_ý"1Ì_ý"150#1ì50#1#1L #1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1ç€&1T ç€&1·R'1˜·R'1_e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1K%ñ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¢â€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1K%ñ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¢â€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1K%ñ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¢â €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1»Û  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ê ê``ÖÖ)L*L4Â4Â>6?6IªIªST^’^’hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttêê``ÖÖ L L  Â88®®$$šš††üür r"è"è$^%^'Ô'Ô)J*J,À,À.6/61¬1¬3"4"6˜6˜89;„;„=ú=ú?p@pBæBæD\E\GÒGÒIFJFLºLºN.O.Q¢Q¢STVŠVŠXþXþZr[r]æ]æ_Z`ZbÎbÎdBeBg¶g¶i*j*lžlžnoq†q†súsúunvnxâxâzV{V}Ê}Ê>€>‚²‚²„&…&‡š‡š‰ŠŒ‚Œ‚ŽöŽö�j‘j“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1¿}!1�¿}!1]¾!1 ]¾!1³Â!1³Â!1‹ç!1€‹ç!1)("1ü)("1…,"1…,"1]Q"1p]Q"1û‘"1ìû‘"1W–"1ôW–"1…Ú"1`…Ú"1åþ"1Ìåþ"1/0#1ì/0#1û~#1L û~#1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1�&1T �&1íR'1˜íR'1•e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–ùÅ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–ùÅ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–ùÅ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Aa  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww ñ ñkkáá)W*W4Í4Í>A?AIµIµS)T)^�^�his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwññkkåå _ _ Ù ÙSSÍÍGGÁÁ;;µµ// © ©"###%™%™'(*…*…,û,û.q/q1ç1ç3]4]6Ó6Ó8I9I;¿;¿=5>5@«@«B!C!E—E—G H J�J�LõLõNiOiQÝQÝSQTQVÅVÅX9Y9[­[­]!^!`•`•b c e}e}gñgñiejelÙlÙnMoMqÁqÁs5t5v©v©xy{‘{‘}~€y€y‚í‚í„a…a‡Õ‡Õ‰IŠIŒ½Œ½Ž1�1‘¥‘¥“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1ï}!1�ï}!1�¾!1 �¾!1GÅ!1GÅ!1Áç!1€Áç!1_("1ü_("1/"1/"1“Q"1p“Q"11’"1ì1’"1ë˜"1ôë˜"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1/0#1ì/0#1û~#1L û~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1�&1T �&1íR'1˜íR'1•e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆ,ø b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1”Ô€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆ,ø b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1”Ô€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆ,ø b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1”Ô €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1»Û  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp ç ç^^ÕÕ)L*L4Ã4Ã>8?8I­I­S"T"^—^—h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppçç^^ÕÕ L L à Ã::±±((ŸŸ�� { {"ò"ò$i%i'à'à)W*W,Î,Î.E/E1¼1¼33436ª6ª8!9!;˜;˜=>@†@†BýBýDtEtGëGëI`J`LÕLÕNJOJQ¿Q¿S4T4V©V©XY[“[“]^`}`}bòbòdgeggÜgÜiQjQlÆlÆn;o;q°q°s%t%všvšxy{„{„}ù}ùn€n‚ã‚ã„X…X‡͇͉BŠBŒ·Œ·Ž,�,‘¡‘¡“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1¹}!1�¹}!1W¾!1 W¾!1³Â!1³Â!1…ç!1€…ç!1#("1ü#("1…,"1…,"1WQ"1pWQ"1õ‘"1ìõ‘"1W–"1ôW–"1‘Ú"1`‘Ú"1ñþ"1Ìñþ"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ó¿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ó¿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ó¿ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Pp  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu í íeeÙÙ)M*M4Á4Á>3?3I¥I¥ST^‰^‰hûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuííeeÝÝ U U Í ÍEE½½55­­%%�� � �"#%y%y'í'í)a*a,Õ,Õ.I/I1½1½31416¥6¥89;�;�=>@u@uBéBéD]E]GÑGÑICJCLµLµN'O'Q™Q™S T V}V}XïXïZa[a]Ó]Ó_E`Eb·b·d)e)g›g›i j llnñnñpcqcsÕsÕuGvGx¹x¹z+{+}�}�€‚�‚�„ó„ó†e‡e‰׉׋IŒI޻޻�-‘-“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1ï}!1�ï}!1�¾!1 �¾!1GÅ!1GÅ!1Áç!1€Áç!1_("1ü_("1/"1/"1“Q"1p“Q"11’"1ì1’"1ë˜"1ôë˜"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1±€&1T ±€&1�R'1˜�R'1)e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆ.ú b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1•Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆ.ú b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1•Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆ.ú b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1•Õ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1»Û  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ç ç]]ÒÒ)G*G4¼4¼>/?/I¢I¢ST^ˆ^ˆhûhûrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqçç]]ÓÓ I I ¿ ¿55««!!——  ƒƒùùo o"å"å$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ì=ì?a@aBÖBÖDKEKGÀGÀI3J3L¦L¦NOQŒQŒSÿSÿUrVrXåXåZX[X]Ë]Ë_>`>b±b±d$e$g—g—i j l}l}nðnðpcqcsÖsÖuIvIx¼x¼z/{/}¢}¢€‚ˆ‚ˆ„û„û†n‡n‰á‰á‹TŒTŽÇŽÇ�:‘:“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1³}!1�³}!1Q¾!1 Q¾!1¹Â!1¹Â!1…ç!1€…ç!1#("1ü#("1‹,"1‹,"1WQ"1pWQ"1õ‘"1ìõ‘"1]–"1ô]–"1ùØ"1`ùØ"1Yý"1ÌYý"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“ö b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1»Û  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æ[[ÐÐ)E*E4¸4¸>+?+IžIžST^„^„h÷h÷rhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  €€õõj j"ß"ß$T%T'É'É)>*>,³,³.(/(1�1�346‡6‡8ü8ü:q;q=ä=ä?W@WBÊBÊD=E=G°G°I#J#L–L–N O Q|Q|SïSïUbVbXÕXÕZH[H]»]»_.`.b¡b¡deg‡g‡iúiúkmlmnànàpSqSsÆsÆu9v9x¬x¬z{}’}’€‚x‚x„ë„ë†\‡\‰Ή΋@Œ@޲޲�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1õ}!1�õ}!1“¾!1 “¾!1GÅ!1GÅ!1Çç!1€Çç!1e("1üe("1/"1/"1™Q"1p™Q"17’"1ì7’"1ë˜"1ôë˜"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1í€&1T í€&1½R'1˜½R'1ee'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘5 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1˜Ø€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘5 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1˜Ø€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘5 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1˜Ø €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1i‰  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp ê ê``ÖÖ)L*L4À4À>4?4I¨I¨ST^�^�hisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppêêddÞÞ X X Ò ÒLLÆÆ@@ºº00¦¦ ’ ’"#%~%~'ô'ô)j*j,à,à.V/V1Ì1Ì3B4B6¸6¸8.9.;¤;¤=>@Œ@ŒBCEtEtGèGèI\J\LÐLÐNDODQ¸Q¸S,T,V V XY[ˆ[ˆ]ü]ü_p`pbäbädXeXgÌgÌi@j@l´l´n(o(qœqœstv„v„xøxøzl{l}à}àT€T‚ȂȄ<…<‡®‡®‰!Š!Œ”Œ”Ž�‘z‘z“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1­}!1�­}!1K¾!1 K¾!1¹Â!1¹Â!1ç!1€ç!1("1ü("1‹,"1‹,"1QQ"1pQQ"1ï‘"1ìï‘"1]–"1ô]–"1ùØ"1`ùØ"1Yý"1ÌYý"1)0#1ì)0#1õ~#1L õ~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%13€&1T 3€&1R'1˜R'1«d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ð¼ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ð¼ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ð¼ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1»Û  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æYYÌÌ)?*?4°4°>!?!I’I’ST^t^thåhårTsT} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  ||ïïb b"Õ"Õ$H%H'»'»).*.,¡,¡./1‡1‡3ú3ú5m6m8à8à:S;S=Ä=Ä?5@5B¦B¦DEGˆGˆIùIùKjLjNÛNÛPLQLS½S½U.V.XŸXŸZ[]�]�_ò_òacbcdÔdÔfEgEi¶i¶k'l'n˜n˜p q szszuëuëw\x\zÍzÍ|>}>¯¯� ‚ „‘„‘†‡‰p‰p‹à‹à�PŽP�À�À’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1õ}!1�õ}!1“¾!1 “¾!1GÅ!1GÅ!1Çç!1€Çç!1e("1üe("1/"1/"1™Q"1p™Q"17’"1ì7’"1ë˜"1ôë˜"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1{€&1T {€&1KR'1˜KR'1ód'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘7 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1™Ù€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘7 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1™Ù€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘7 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1™Ù €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1t”  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp è è\\ÐÐ)D*D4¶4¶>(?(IšIšS T ^~^~hðhðr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppèè``ØØ P P È È@@¸¸00¨¨�� x x"ì"ì$`%`'Ô'Ô)H*H,¼,¼.0/01¤1¤346Œ6Œ89;t;t=æ=æ?X@XBÊBÊD[>]°]°_"`"b”b”degxgxiêiêk\l\nÎnÎp@q@s²s²u$v$x–x–z{}z}zìì�^‚^„ЄІ@‡@‰±‰±‹"Œ"ޓޓ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1­}!1�­}!1K¾!1 K¾!1¹Â!1¹Â!1ç!1€ç!1("1ü("1‹,"1‹,"1QQ"1pQQ"1ï‘"1ìï‘"1]–"1ô]–"1óØ"1`óØ"1Sý"1ÌSý"1)0#1ì)0#1õ~#1L õ~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ó¿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ó¿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ó¿ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1»Û  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo ç ç\\ÑÑ)F*F4»4»>/?/I£I£ST^‹^‹hÿhÿrsss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooçç__×× O O Ç Ç??··//§§‘‘ { {"ð"ð$e%e'Ú'Ú)O*O,Ä,Ä.9/91®1®3#4#6˜6˜8 9 ;‚;‚=÷=÷?l@lBáBáDVEVGËGËI?J?L³L³N'O'Q›Q›STVƒVƒX÷X÷Zk[k]ß]ß_S`SbÇbÇd;e;g¯g¯i#j#l—l—n o qqsósóugvgxÛxÛzO{O}Ã}Ã7€7‚«‚«„…‡“‡“‰ŠŒ}Œ}ŽòŽò�g‘g“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1û}!1�û}!1™¾!1 ™¾!1AÅ!1AÅ!1Çç!1€Çç!1e("1üe("1/"1/"1™Q"1p™Q"17’"1ì7’"1å˜"1ôå˜"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1)0#1ì)0#1õ~#1L õ~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1;�&1T ;�&1 S'1˜ S'1³e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš>  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1”Ô€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš>  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1”Ô€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš>  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1”Ô €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1nŽ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll å å^^××)P*P4É4É>@?@I·I·S,T,^Ÿ^Ÿhisƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllåå^^×× P P É ÉBB»»44­­&&ŸŸ ‘ ‘" # %ƒ%ƒ'ü'ü)u*u,î,î.g/g1à1à3Y4Y6Ò6Ò8K9K;Ä;Ä==>=@¶@¶B/C/E¨E¨G!H!J˜J˜LMO†O†QýQýStTtVëVëXbYb[Ù[Ù]P^P`Ç`Çb*?*I�I�ST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““}}òòg g"Ü"Ü$Q%Q'Æ'Æ);*;,°,°.%/%1š1š346„6„8ù8ù:n;n=ã=ã?X@XBÍBÍDBEBG·G·I*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b§b§deg‹g‹iýiýkolonánápSqSsÅsÅu7v7x©x©z{}�}�ÿÿ�q‚q„ã„ã†U‡U‰ȉȋ;Œ;ޮޮ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1û}!1�û}!1™¾!1 ™¾!1AÅ!1AÅ!1Íç!1€Íç!1k("1ük("1/"1/"1ŸQ"1pŸQ"1=’"1ì=’"1å˜"1ôå˜"1»Ú"1`»Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš@  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1•Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš@  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1•Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš@  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1•Õ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ÿ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ã ãZZÑÑ)H*H4¿4¿>4?4I©I©ST^�^�hþhþrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllããZZÑÑ H H ¿ ¿66­­$$››‰‰ w w"î"î$e%e'Ü'Ü)S*S,Ê,Ê.A/A1¸1¸3/4/6¦6¦89;”;”= > @‚@‚BùBùDpEpGçGçI\J\LÑLÑNFOFQ»Q»S0T0V¥V¥XY[�[�]^`y`ybìbìd_e_gÒgÒiEjEl¸l¸n)o)qšqšs t v|v|xíxíz^{^}Ï}Ï@€@‚±‚±„"…"‡’‡’‰ŠŒtŒtŽåŽå�V‘V“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1§}!1�§}!1E¾!1 E¾!1¹Â!1¹Â!1yç!1€yç!1("1ü("1‹,"1‹,"1KQ"1pKQ"1é‘"1ìé‘"1]–"1ô]–"1óØ"1`óØ"1Sý"1ÌSý"1#0#1ì#0#1ï~#1L ï~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1ù€&1T ù€&1ÉR'1˜ÉR'1qe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ð¼ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ð¼ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ð¼ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1»Û  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk â âYYÐÐ)G*G4¾4¾>5?5I¬I¬S#T#^•^•hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkââYYÐÐ G G ¾ ¾55¬¬##ššˆˆÿÿv v"í"í$d%d'Û'Û)R*R,É,É.@/@1·1·3.4.6¥6¥89;“;“= > @�@�BøBøDoEoGæGæI]J]LÔLÔNKOKQÂQÂS9T9V°V°X'Y'[ž[ž]^`Œ`ŒbcezezgñgñihjhlßlßnQoQqÃqÃs5t5v§v§xy{‹{‹}ý}ýo€o‚á‚á„S…S‡Ňʼn7Š7Œ©Œ©Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1~!1�~!1Ÿ¾!1 Ÿ¾!1;Å!1;Å!1Íç!1€Íç!1k("1ük("1 /"1 /"1ŸQ"1pŸQ"1=’"1ì=’"1ߘ"1ôߘ"1�Ú"1`�Ú"1ýþ"1Ìýþ"1#0#1ì#0#1ï~#1L ï~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1A�&1T A�&1S'1˜S'1¹e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£G b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1–Ö€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£G b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1–Ö€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£G b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1–Ö €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1g‡  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àXXÐÐ)H*H4À4À>6?6I¬I¬S"T"^˜^˜his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  �� € €"ø"ø$p%p'è'è)`*`,Ø,Ø.P/P1È1È3@4@6¸6¸8090;¨;¨= > @˜@˜BCEˆEˆGHJvJvLìLìNbObQØQØSNTNVÄVÄX:Y:[°[°]&^&`œ`œbceˆeˆgþgþitjtlêlên`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¡}!1�¡}!1?¾!1 ?¾!1¹Â!1¹Â!1sç!1€sç!1("1ü("1‹,"1‹,"1EQ"1pEQ"1ã‘"1ìã‘"1]–"1ô]–"1óØ"1`óØ"1Sý"1ÌSý"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ê¶ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ê¶ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ê¶ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1»Û  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii Þ ÞSSÈÈ)=*=4²4²>'?'IœIœST^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiÞÞSSÈÈ = = ² ²''œœ††ûûppååZ Z"Ï"Ï$D%D'¹'¹).*.,£,£./1�1�346w6w8ì8ì:a;a=Ö=Ö?K@KBÀBÀD5E5GªGªIJL”L”N O Q~Q~SóSóUhVhXÝXÝZR[R]Ç]Ç_<`*?*IžIžST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽzzððf f"Ü"Ü$R%R'È'È)>*>,´,´.*/*1 1 346Œ6Œ89;x;x=î=î?d@dBÚBÚDPEPGÆGÆI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1¡}!1�¡}!1?¾!1 ?¾!1­Â!1­Â!1sç!1€sç!1("1ü("1,"1,"1EQ"1pEQ"1ã‘"1ìã‘"1Q–"1ôQ–"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠí¹ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠí¹ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠí¹ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1»Û  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk å å__ÕÕ)K*K4Á4Á>5?5I©I©ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkåå__ÙÙ S S Í ÍGGÁÁ;;µµ//©©## � �"#%�%�'(*y*y,ï,ï.e/e1Û1Û3Q4Q6Ç6Ç8=9=;³;³=)>)@Ÿ@ŸBCE‹E‹GHJuJuLéLéN]O]QÑQÑSETEV¹V¹X-Y-[¡[¡]^`‰`‰býbýdqeqgågåiYjYlÍlÍnAoAqµqµs)t)v�v�xy{…{…}ù}ùm€m‚á‚á„U…U‡ɇɉ=Š=Œ±Œ±Ž%�%‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1~!1�~!1Ÿ¾!1 Ÿ¾!15Å!15Å!1Óç!1€Óç!1q("1üq("1/"1/"1¥Q"1p¥Q"1C’"1ìC’"1Ù˜"1ôÙ˜"1‘Ú"1`‘Ú"1ñþ"1Ìñþ"10#1ì0#1é~#1L é~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬P b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ѐ} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬P b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ѐ} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬P b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ð €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1b‚  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>,?,I¡I¡ST^‹^‹hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbæbæd[e[gÐgÐiEjElºlºn/o/q¤q¤stvŽvŽxy{x{x}í}íb€b‚ׂׄL…L‡Á‡Á‰6Š6Œ«Œ«Ž � ‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1›}!1�›}!19¾!1 9¾!1­Â!1­Â!1mç!1€mç!1 ("1ü ("1,"1,"1?Q"1p?Q"1Ý‘"1ìÝ‘"1Q–"1ôQ–"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ç³ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 L€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ç³ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 L€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ç³ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 L €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1»Û  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áYYÍÍ)A*A4µ4µ>'?'I™I™S T ^}^}hïhïrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‘‘  � �"ù"ù$m%m'á'á)U*U,É,É.=/=1±1±3%4%6™6™8 9 ;�;�=õ=õ?i@iBÝBÝDQEQGÅGÅI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1~!1�~!1¥¾!1 ¥¾!15Å!15Å!1Óç!1€Óç!1q("1üq("1/"1/"1¥Q"1p¥Q"1C’"1ìC’"1Ù˜"1ôÙ˜"1»Ú"1`»Ú"1ÿ"1Ìÿ"10#1ì0#1é~#1L é~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬R b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‘Ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬R b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‘Ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬R b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‘Ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1‡§  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÆÆ);*;4°4°>#?#I–I–S T ^|^|hïhïrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnänäpWqWsÊsÊu=v=x°x°z#{#}–}– € ‚|‚|„ï„ï†b‡b‰Õ‰Õ‹HŒH޻޻�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1›}!1�›}!19¾!1 9¾!1­Â!1­Â!1gç!1€gç!1("1ü("1,"1,"19Q"1p9Q"1ב"1ìב"1Q–"1ôQ–"1CÚ"1`CÚ"1£þ"1Ì£þ"10#1ì0#1é~#1L é~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ê¶ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ê¶ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ê¶ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations16  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94¬4¬>?I’I’ST^z^zhíhír^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ø=Ø?K@KB¾B¾D1E1G¤G¤IJLŠLŠNýNýPpQpSãSãUVVVXÉXÉZ<[<]¯]¯_"`"b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†R‡R‰ĉċ6Œ6ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1~!1�~!1¥¾!1 ¥¾!1/Å!1/Å!1Ùç!1€Ùç!1w("1üw("1/"1/"1«Q"1p«Q"1I’"1ìI’"1Ó˜"1ôÓ˜"1çØ"1`çØ"1Gý"1ÌGý"10#1ì0#1é~#1L é~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µY% b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1“Ó€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µY% b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1“Ó€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µY% b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1“Ó €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1»Û  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Þ ÞTTÊÊ)@*@4´4´>(?(IœIœST^Ž^Žhistst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÞÞXXÒÒ L L Æ Æ@@ºº44®®$$šš † †"ü"ü$r%r'è'è)^*^,Ô,Ô.J/J1À1À36466¬6¬8"9";˜;˜= > @€@€BôBôDhEhGÜGÜIPJPLÄLÄN8O8Q¬Q¬S T V”V”XY[|[|]ð]ð_d`dbâbâdVeVgÊgÊi>j>l²l²n&o&qšqšstv‚v‚xöxözj{j}Þ}ÞR€R‚ƂƄ:…:‡¬‡¬‰ŠŒ’Œ’Ž�‘x‘x“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1•}!1�•}!13¾!1 3¾!1­Â!1­Â!1aç!1€aç!1ÿ'"1üÿ'"1,"1,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1Q–"1ôQ–"1aÚ"1`aÚ"1Áþ"1ÌÁþ"10#1ì0#1é~#1L é~#1™$1¬ ™$1qò$1p qò$1«Þ%1t «Þ%1€&1T €&1ëQ'1˜ëQ'1“d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒä° b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒä° b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒä° b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations13S  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚMMÀÀ)3*34¤4¤>?I†I†SûSû]l^lhÝhÝrLsL} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠýýppããV V"É"É$<%<'¯'¯)"*",•,•./1{1{3î3î5a6a8Ô8Ô:G;G=¸=¸?)@)BšBšD E G|G|IíIíK^L^NÏNÏP@Q@S±S±U"V"X“X“Z[]u]u_æ_æa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1 ~!1� ~!1«¾!1 «¾!1/Å!1/Å!1Ùç!1€Ùç!1w("1üw("1/"1/"1«Q"1p«Q"1I’"1ìI’"1Ó˜"1ôÓ˜"1çØ"1`çØ"1Gý"1ÌGý"10#1ì0#1é~#1L é~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1c€&1T c€&13R'1˜3R'1Ûd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µ[' b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1’Ò€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µ[' b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1’Ò€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µ[' b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1’Ò €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¼Ü  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜPPÄÄ)8*84ª4ª>?IŽIŽST^z^zhìhìr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœ„„øøl l"à"à$T%T'È'È)<*<,°,°.$/$1˜1˜3 4 6€6€8ô8ô:h;h=Ú=Ú?L@LB¾B¾D0E0G¢G¢IJL†L†NøNøPjQjSÜSÜUNVNXÀXÀZ2[2]¤]¤_`b�b�degtgtiæiækXlXnÊnÊp%?%I™I™ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››……úúo o"ä"ä$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`IbÉbÉd=e=g±g±i%j%l™l™n o q�q�sõsõuivixÝxÝzQ{Q}Å}Å9€9‚­‚­„!…!‡•‡•‰ Š ŒŒŽôŽô�i‘i“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1 ~!1� ~!1«¾!1 «¾!1)Å!1)Å!1ßç!1€ßç!1}("1ü}("1û."1û."1±Q"1p±Q"1O’"1ìO’"1͘"1ô͘"1Ú"1`Ú"1sþ"1Ôsþ"10#1ô0#1ã~#1T ã~#1Õ$1´ Õ$1ýð$1x ýð$1ÛÀ%1| ÛÀ%15�&1\ 5�&1+Q'1 +Q'1Óc'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оb. b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Í€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оb. b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Í€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оb. b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Í €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1§ïW   bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ù ÙRRÕÕ)N*N4Ç4Ç>>?>IµIµS*T*^©^©his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÙÙRRËË D D ½ ½66¯¯((¡¡““  … …"þ"þ$�%�'ú'ú)s*s,ì,ì.e/e1Þ1Þ3W4W6Ð6Ð8I9I;Â;Â=;>;@´@´B-C-E¦E¦GHJ–J–L M O„O„QûQûSrTrVéVéX`Y`[×[×]N^N`Å`Åb:c:e¯e¯g$h$j™j™lmo�o�qrtstsvævæxYyY{Ì{Ì}?~?€²€²‚%ƒ%…˜…˜‡ ˆ Š{Š{ŒíŒíŽ_�_‘ёѓ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1�}!1��}!1-¾!1 -¾!1¡Â!1¡Â!1[ç!1€[ç!1ù'"1üù'"1s,"1s,"1-Q"1p-Q"1Ë‘"1ìË‘"1E–"1ôE–"1ÃØ"1`ÃØ"1#ý"1Ì#ý"10#1ì0#1ã~#1L ã~#1±$1¬ ±$1Ùð$1p Ùð$1·À%1t ·À%1‡€&1T ‡€&1}P'1˜}P'1%c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€á­ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 I€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€á­ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 I€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1L€á­ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 I €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1»·  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×LLÅÅ):*:4¯4¯>"?"I•I•ST^…^…h÷h÷risi} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡üüqqææ[ ["Ð"Ð$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I"J"L•L•NOQ{Q{SîSîUaVaXÔXÔZG[G]º]º_-`-b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰̉̋?Œ?޲޲�%‘%“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1 ~!1� ~!1«¾!1 «¾!1)Å!1)Å!1ßç!1€ßç!1}("1ü}("1û."1û."1±Q"1p±Q"1O’"1ìO’"1͘"1ô͘"1=Ú"1`=Ú"1�þ"1Ô�þ"10#1ô0#1ã~#1T ã~#1Ã$1´ Ã$1ëð$1x ëð$1ÉÀ%1| ÉÀ%1½€&1\ ½€&1³P'1 ³P'1[c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оd0 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ŽÎ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оd0 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ŽÎ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оd0 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ŽÎ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1³û3  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` × ×NNÍÍ)D*D4»4»>0?0I¥I¥ST^•^•hisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``××NNÅÅ < < ³ ³**¡¡��}}ôôk k"â"â$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n1o1q¢q¢stv„v„xõxõzf{f}×}×H€H‚¹‚¹„*…*‡š‡š‰ Š Œ|Œ|ŽíŽí�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„+i!1S}!1S}!1ñ½!1|ñ½!1UÃ!1„UÃ!1‹Æ!1ð‹Æ!1Ð!1|Ð!1KÓ!1HKÓ!1%ç!1Ô%ç!1Ã'"1PÃ'"1'-"1X'-"1]0"1Ä]0"1ç9"1Pç9"1="1="1÷P"1¨÷P"1•‘"1$ •‘"1ù–"1, ù–"1/š"1˜ /š"1¹£"1$ ¹£"1ï¦"1ð ï¦"11Ô"1| 1Ô"1mÚ"1XmÚ"1Íþ"1ÔÍþ"1,#1ô,#1O:#1ÐO:#1 o#1\ o#1é�#1€é�#1ç˜#1¬ç˜#1Õ$18Õ$1­ò$1ü­ò$1çÞ%1çÞ%1aì%1àaì%1/�&1l/�&1ÿR'1°ÿR'1§e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Pµ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1½ý bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Pµ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1½ý  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Pµ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1½ý bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 |œHh&F bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1HM�Úg§ô4  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dˆMÍÁAšzúçg3³´4�NΛèhµ5‚OÏ œ é i ¶ 6 ƒ  ž Þ P Ð  ] ª ê 7wÄ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` © ©RRÕÕ)N*N4Ç4Ç>@?@I·I·S,T,^«^«his‘s‘} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÙÙRRËË D D � �øøqq““  … …"þ"þ$�%�'ú'ú)s*s,ì,ì.e/e1Þ1Þ3W4W6Ð6Ð8I9I;Â;Â=;>;@´@´B-C-E¦E¦GHJ˜J˜LMO†O†QýQýStTtVëVëXbYb[Ù[Ù]P^P`Ç`Çb3?3I§I§ST^›^›hisƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯))££ — —"#%‰%‰'ÿ'ÿ)u*u,ë,ë.a/a1§1§346“6“8 9 ;;=%>%@›@›BCE‡E‡GýGýIsJsLçLçN[O[QÏQÏSCTCV·V·X+Y+[Ÿ[Ÿ]^`‡`‡bûbûdoeogãgãiWjWlËlËnKoKq¿q¿s3t3v§v§xy{�{�}~€w€w‚ë‚ë„_…_‡Ó‡Ó‰GŠGŒ»Œ»Ž/�/‘£‘£“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„+i!1S}!1S}!1ñ½!1|ñ½!1UÃ!1„UÃ!1‹Æ!1ð‹Æ!1Ð!1|Ð!1KÓ!1HKÓ!1%ç!1Ô%ç!1Ã'"1PÃ'"1'-"1X'-"1]0"1Ä]0"1ç9"1Pç9"1="1="1÷P"1¨÷P"1•‘"1$ •‘"1ù–"1, ù–"1/š"1˜ /š"1¹£"1$ ¹£"1ï¦"1ð ï¦"11Ô"1| 1Ô"17Ú"1X7Ú"1—þ"1Ô—þ"1Ù+#1ôÙ+#1Ó:#1ÐÓ:#1�o#1\�o#1}�#1€}�#1{›#1¬{›#1½$18½$1•ò$1ü•ò$1ÏÞ%1ÏÞ%1Iì%1àIì%1½€&1l½€&1�R'1°�R'15e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P³ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1½ý bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P³ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1½ý  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P³  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1½ý bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 zš7$ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1o¯�Á^Ë x¸  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1doï^ÞMÍ<¼ìl¥%+«`àš› ‰øxåeÒR ¿ ? ¬ , ™   W †  s ó `àO�þ>«ëX˜ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` ¦ ¦NNÍÍ)D*D4»4»>2?2I§I§ST^—^—hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``××NNÅÅ < < ‚ ‚ùùppçç^^}}ôôk k"â"â$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbêbêd]e]gÐgÐiCjCl¶l¶n3o3q¤q¤stv†v†x÷x÷zh{h}Ù}ÙJ€J‚»‚»„,…,‡ž‡ž‰ŠŒ€Œ€ŽñŽñ�b‘b“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1+~!1è+~!1ɾ!1dɾ!1qÅ!1lqÅ!1ûÎ!1èûÎ!11Ò!1´1Ò!1÷ç!1@÷ç!1•("1¼•("1C/"1ÄC/"1Í8"1@Í8"1<"1 <"1ÉQ"1˜ÉQ"1g’"1g’"1™"1 ™"1Ÿ¢"1˜ Ÿ¢"1Õ¥"1d Õ¥"1Ó"1ð Ó"1mÚ"1Ì mÚ"1Íþ"1H Íþ"1,#1h,#1O:#1DO:#1 o#1Ð o#1é�#1ôé�#1ç˜#1 ç˜#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1‡€&1à‡€&1WR'1$WR'1ÿd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¨Ü„P b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M�†ÆÚ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¨Ü„P b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M�†ÆÚ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¨Ü„P  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M�†ÆÚ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �¯½ÝNn bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1M��ÍÛh¨õ5  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dMÍšçg´4 Š�ÉINΈ›GÇèhµ5‚OÏ œ é i ¶ 6 ƒ  P Ð ~ ¾  ] ª ê 7wÄ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSËË)*4³4³>'?'I™I™S T ^‰^‰hûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$k%k'ß'ß)S*S,Ç,Ç.;/;1~1~3ò3ò5f6f8Ú8Ú:N;N=ó=ó?g@gBÛBÛDOEOGÃGÃI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnñnñpcqcsÕsÕuGvGx¹x¹z+{+}�}�€‚�‚�„ó„ó†e‡e‰׉׋IŒI޻޻�-‘-“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1�}!1è�}!1-¾!1d-¾!1wÅ!1lwÅ!1Ï!1èÏ!17Ò!1´7Ò!1aç!1@aç!1ÿ'"1¼ÿ'"1I/"1ÄI/"1Ó8"1@Ó8"1 <"1  <"13Q"1˜3Q"1Ñ‘"1Ñ‘"1™"1 ™"1¥¢"1˜ ¥¢"1Û¥"1d Û¥"1Ó"1ð Ó"1‹Ú"1Ì ‹Ú"1ëþ"1H ëþ"1-,#1h-,#1U:#1DU:#1o#1Ðo#1ï�#1ôï�#1í˜#1 í˜#1Ï$1¬Ï$1§ò$1p§ò$1áÞ%1táÞ%1[ì%1T[ì%15�&1à5�&1S'1$S'1­e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�æ² b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TM�Ú bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�æ² b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TM�Ú bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�æ²  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TM�Ú  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �¡,Lg‡ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1 KM�Úg§ô4  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d ‹MÍÆFš�çg<¼´4�NΛèhµ5‚OÏ œ é i ¶ 6 ƒ  » û P Ð  ] ª ê 7wÄ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ø ØPPÈÈ)@*@4¸4¸>0?0IvIvST^œ^œhisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ØØPPÈÈ @ @ ¸ ¸00¨¨  ˜˜ˆˆ x x"ð"ð$h%h'à'à)X*X,Ð,Ð.H/H1À1À38486°6°8(9(; ; =>@�@�BCE€E€GøGøIpJpLæLæN\O\QÒQÒSHTHVŽVŽXY[z[z]ð]ð_f`fb c e‚e‚gøgøinjnläländodqÚqÚsPtPvÆvÆxpð{ûhè8¸`àXØPÐHÈ@À:º4 ´ . ® ( ¨ " ¢ Ñ   œ –�HÀzº4t bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØMMÂÂ)7*74¬4¬>?I’I’ST^J^Jhíhír`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?K@KB¾B¾D1E1G¤G¤IJLŠLŠNýNýPpQpSãSãUVVVXÉXÉZ<[<]¯]¯_"`"b—b—d e g}g}iðiðkclcn¦n¦pqsŒsŒuÿuÿwrxrz{}ˆ}ˆûû�n‚n„á„á†T‡T‰ƉƋ8Œ8ŽªŽª�‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1�}!1è�}!1-¾!1d-¾!1wÅ!1lwÅ!1Ï!1èÏ!17Ò!1´7Ò!1[ç!1@[ç!1ù'"1¼ù'"1I/"1ÄI/"1Ó8"1@Ó8"1 <"1  <"1-Q"1˜-Q"1Ë‘"1Ë‘"1™"1 ™"1¥¢"1˜ ¥¢"1Û¥"1d Û¥"1Ó"1ð Ó"1aÚ"1Ì aÚ"1Áþ"1H Áþ"1,#1h,#1U:#1DU:#1o#1Ðo#1ï�#1ôï�#1í˜#1 í˜#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1½€&1ཀ&1�R'1$�R'15e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚å± b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 SM�Ú bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚å± b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 SM�Ú bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚å±  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 SM�Ú  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 {›:Z& bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1M�†ÆÚg§ô4  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dMÍšçg´4÷w�²2NÎmí›(¨èhµ5‚OÏ œ é i ¶ 6 ƒ  P Ð 6 v  ] ª ê 7wÄ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ö ÖLLÂÂ)8*84®4®>$?$IgIgS T ^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††üürrèè^ ^"Ô"Ô$J%J'À'À)6*6,¬,¬."/"1˜1˜346„6„8ú8ú:p;p=æ=æ?\@\BÒBÒDHEHG¾G¾I4J4L¨L¨NOQ�Q�STVGVGX»X»Z/[/]£]£_`b¼b¼d0e0g¤g¤ijlŒlŒnoq|q|sðsðudvdxØxØzL{L}À}À4€4‚¨‚¨„…‡�‡�‰ŠŒxŒxŽìŽì�`‘`“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1~!1è~!1½¾!1d½¾!1wÅ!1lwÅ!1Ï!1èÏ!17Ò!1´7Ò!1ñç!1@ñç!1�("1¼�("1I/"1ÄI/"1Ó8"1@Ó8"1 <"1  <"1ÃQ"1˜ÃQ"1a’"1a’"1™"1 ™"1¥¢"1˜ ¥¢"1Û¥"1d Û¥"1Ó"1ð Ó"1sÚ"1Ì sÚ"1Óþ"1H Óþ"1,#1h,#1©:#1D©:#1eo#1Ðeo#1“�#1ô“�#1‘š#1 ‘š#1“$1¬“$1kò$1pkò$1¥Þ%1t¥Þ%1ì%1Tì%1€&1à€&1ëQ'1$ëQ'1“d'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�ÑyE b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M�žÞÚ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�ÑyE b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M�žÞÚ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�ÑyE  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M�žÞÚ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ’²'GPp bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1c£~¾F§çHˆ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dcãFÆ)© Œõuïo²2oïÒR,¬µ5˜yùZÚ; »  œ ý } Þ ^ N Ž ¿ ?   ƒÃ&fÇh¨ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØKK¾¾)1*14¤4¤>?I†I†SûSû];^;hÝhÝrNsN} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆûûnnááT T"Ç"Ç$:%:'­'­) * ,“,“./1y1y3ì3ì5_6_8Ò8Ò:E;E=¸=¸?)@)BšBšD E G|G|IíIíK^L^NÏNÏP@Q@S±S±U"V"X“X“Z[]u]u_æ_æa[b[dÌdÌf=g=i®i®kln_n_pÐpÐrAsAu²u²w#x#zÅzÅ|6}6§§�‚„‰„‰†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1•}!1Е}!13¾!1L3¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1aç!1(aç!1ÿ'"1¤ÿ'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"13Q"1€3Q"1Ñ‘"1üÑ‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1c2#1,c2#1ûŠ#1˜ûŠ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1ó€&1Ìó€&1ÃR'1ÃR'1ke'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ç³ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TM�Ú bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ç³ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TM�Ú bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1N‚ç³  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TM�Ú  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 †¦2RjŠ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1„ÁA~þ;»Ô   €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` × ×NNÅÅ)<*<4³4³>*?*IŸIŸST^‰^‰hþhþrsss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``××NNÅÅ < < ³ ³**¡¡��}}ôôk k"â"â$Y%Y'Ð'Ð)G*G,¾,¾.5/51¬1¬3#4#6š6š89;ˆ;ˆ=ÿ=ÿ?v@vBíBíDdEdGÛGÛIRJRLÇLÇN'?'I›I›ST^ƒ^ƒh÷h÷rmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��‡‡üüq q"æ"æ$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡�‡�‰ŠŒwŒwŽìŽì�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1•}!1Е}!13¾!1L3¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1aç!1(aç!1ÿ'"1¤ÿ'"1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"13Q"1€3Q"1Ñ‘"1üÑ‘"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1ß4#1,ß4#1y‹#1˜y‹#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒæ² b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UM�Ú bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒæ² b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UM�Ú bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒæ²  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UM�Ú  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 |œOo( bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Á€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1øxµ5rò/¯Q ‘  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×MMÂÂ)7*74¬4¬>!?!I”I”ST^z^zhíhír`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b£b£deg‰g‰iüiükolonânâpUqUsÈsÈu;v;x®x®z!{!}”}”€‚z‚z„í„í†`‡`‰Ó‰Ó‹FŒF޹޹�,‘,“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1~!1Ð~!1·¾!1L·¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1ëç!1(ëç!1‰("1¤‰("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"1½Q"1€½Q"1[’"1ü[’"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1å4#1,å4#1Œ#1˜Œ#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1{€&1Ì{€&1KR'1KR'1ód'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Æn: b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M�”Ôß bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Æn: b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M�”Ôß bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Æn:  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M�”Ôß  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 •µJjRr bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1‡Ç€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1œãcª*qñç '  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84­4­>"?"I•I•ST^y^yhëhër_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?N@NBÃBÃD8E8G­G­I"J"L•L•NOQ{Q{SîSîUaVaXÔXÔZG[G]º]º_-`-bŸbŸdegƒgƒiõiõkglgnÙnÙpKqKs½s½u/v/x¡x¡z{}…}…÷÷�i‚i„Û„Û†O‡O‰‰‹5Œ5ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1›}!1Л}!19¾!1L9¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1gç!1(gç!1("1¤("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"19Q"1€9Q"1ב"1üב"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1aÚ"1´ aÚ"1Áþ"10 Áþ"1,#1P,#1c2#1,c2#1‹#1˜‹#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Û€&1ÌÛ€&1«R'1«R'1Se'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒè´ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IO�à  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒè´ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IO�à  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒè´  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IO�à   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ½Ý‹«8X bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÄDˆLÌP �  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Þ ÞTTÊÊ)@*@4¶4¶>*?*IžIžST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÞÞXXÒÒ L L Æ Æ@@ºº44®®$$šš † †"ü"ü$r%r'è'è)^*^,Ô,Ô.J/J1À1À36466¬6¬8"9";˜;˜=>@‚@‚BöBöDjEjGÞGÞIRJRLÆLÆN:O:Q®Q®S"T"V–V–X Y [~[~]ò]ò_f`fbÚbÚdNeNgÂgÂi6j6lªlªnoq’q’stvzvzxîxîzb{b}Ö}ÖJ€J‚¾‚¾„2…2‡¦‡¦‰ŠŒŒŒŒŽÿŽÿ�r‘r“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1 ~!1Ð ~!1«¾!1L«¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ßç!1(ßç!1}("1¤}("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1±Q"1€±Q"1O’"1üO’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1µÚ"1´ µÚ"1ÿ"10 ÿ"1W,#1PW,#1c2#1,c2#1?Œ#1˜?Œ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1í€&1Ìí€&1½R'1½R'1ee'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹_+ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R’”Ôî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹_+ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R’”Ôî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹_+  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R’”Ôî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x˜ @„¤ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÒR¤$vö P  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>1?1I¨I¨ST^‘^‘hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIYJYLÐLÐNGOGQ¾Q¾S5T5V¬V¬X#Y#[š[š]^`ˆ`ˆbÿbÿdvevgígíidjdlÛlÛnMoMq¿q¿s1t1v£v£xy{‡{‡}ù}ùk€k‚݂݄O…O‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1•}!1Е}!13¾!1L3¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1gç!1(gç!1("1¤("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"19Q"1€9Q"1ב"1üב"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"11Ú"1´ 1Ú"1‘þ"10 ‘þ"1Ó+#1PÓ+#1å4#1,å4#1Œ#1˜Œ#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1c€&1Ìc€&13R'13R'1Ûd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„ç³ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 HNŽÞ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„ç³ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 HNŽÞ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„ç³  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 HNŽÞ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ½Ý}� * bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1„Ä€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�ÔT˜\ÜÔ   €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜPPÄÄ)8*84¬4¬>?I�I�ST^t^thæhærXsX} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœ„„øøl l"à"à$T%T'È'È)<*<,°,°.$/$1˜1˜3 4 6€6€8ô8ô:h;h=Ü=Ü?N@NBÀBÀD2E2G¤G¤IJLˆLˆNúNúPlQlSÞSÞUPVPXÂXÂZ4[4]¦]¦_`bŠbŠdüdüfngniàiàkRlRnÄnÄp6q6s¨s¨uvxŒxŒzþzþ|p}pââ�T‚T„ƄƆ8‡8‰©‰©‹ŒŽ‹Ž‹�ü�ü’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1~!1Ð~!1±¾!1L±¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1åç!1(åç!1ƒ("1¤ƒ("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1·Q"1€·Q"1U’"1üU’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1�Ú"1´ �Ú"1ýþ"10 ýþ"1?,#1P?,#1ñ4#1,ñ4#1ÏŒ#1˜ÏŒ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»c/ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S““Óð0 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»c/ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S““Óð0 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»c/  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S““Óð0  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ˜¸o�Tt bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1’Ò€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1HÈšìl¾>¢ â  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ø ØMMÂÂ)7*74¬4¬>#?#I˜I˜S T ^^hñhñrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccØØMM 7 7 ¬ ¬!!––  €€õõjjßßT T"É"É$>%>'³'³)(*(,�,�./1‡1‡3ü3ü5q6q8æ8æ:[;[=Ð=Ð?E@EBºBºD/E/G¤G¤IJL�L�NOQzQzSïSïUdVdXÙXÙZN[N]Ã]Ã_8`8b­b­d"e"g—g—i j l�l�nónópeqes×s×uIvIx»x»z-{-}Ÿ}Ÿ€‚ƒ‚ƒ„õ„õ†g‡g‰ىًKŒK޽޽�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1›}!1Л}!19¾!1L9¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1mç!1(mç!1 ("1¤ ("1y/"1¬y/"19"1(9"19<"1ô9<"1?Q"1€?Q"1Ý‘"1üÝ‘"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1mÚ"1´ mÚ"1Íþ"10 Íþ"1,#1P,#1c2#1,c2#13Œ#1˜3Œ#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1/�&1Ì/�&1ÿR'1ÿR'1§e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„éµ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Ie¥@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„éµ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Ie¥@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„éµ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Ie¥@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ½Ý�°>^ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÏOžmí @  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ý ÝVVÏÏ)H*H4Á4Á>:?:I±I±S&T&^™^™h i ss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÝÝVVÏÏ H H Á Á::³³,,¥¥—— ‰ ‰"#%{%{'ô'ô)m*m,æ,æ._/_1Ø1Ø3Q4Q6Ê6Ê8C9C;¼;¼=5>5@®@®B'C'E E GHJ’J’L M O€O€Q÷Q÷SnTnVåVåX\Y\[Ó[Ó]J^J`Á`Áb6c6e«e«g h j•j•l m o}o}qðqðsctcvÖvÖxIyI{¼{¼}/~/€¢€¢‚ƒ…ˆ…ˆ‡û‡û‰mŠmŒߌߎQ�Q‘ÑÓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1~!1Ð~!1¥¾!1L¥¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Ùç!1(Ùç!1w("1¤w("1[/"1¬[/"1å8"1(å8"1<"1ô<"1«Q"1€«Q"1I’"1üI’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1¯Ú"1´ ¯Ú"1ÿ"10 ÿ"1Q,#1PQ,#1c2#1,c2#1óŒ#1˜óŒ#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1�&1Ì�&1ÕR'1ÕR'1}e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®T  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 R’‰Éâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®T  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 R’‰Éâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®T   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 R’‰Éâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 {›&FŸ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€Ý]º:—À  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii ã ã]]ÓÓ)I*I4¿4¿>5?5I©I©ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiãã]]×× Q Q Ë ËEE¿¿99³³--§§!! › ›"#%‹%‹'(*w*w,í,í.c/c1Ù1Ù3O4O6Å6Å8;9;;±;±='>'@�@�BCE‰E‰GÿGÿIuJuLéLéN]O]QÑQÑSETEV¹V¹X-Y-[¡[¡]^`‰`‰býbýdqeqgågåiYjYlÍlÍnAoAqµqµs)t)v�v�xy{…{…}ù}ùm€m‚á‚á„U…U‡ɇɉ=Š=Œ±Œ±Ž%�%‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1›}!1Л}!19¾!1L9¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1mç!1(mç!1 ("1¤ ("1y/"1¬y/"19"1(9"19<"1ô9<"1?Q"1€?Q"1Ý‘"1üÝ‘"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"17Ú"1´ 7Ú"1—þ"10 —þ"1Ù+#1PÙ+#1÷4#1,÷4#1ÃŒ#1˜ÃŒ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…è´ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jd¤þ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…è´ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jd¤þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…è´  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jd¤þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ½Ý~ž , bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Ï€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1<¼ ‹ÚZ©)� Ï  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I£I£ST^‡^‡høhørjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbæbædYeYgÌgÌi?j?l²l²n#o#q”q”stvvvvxçxçzX{X}É}É:€:‚«‚«„…‡އމÿ‰ÿ‹pŒpŽáŽá�R‘R“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1 ~!1Ð ~!1«¾!1L«¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1ßç!1(ßç!1}("1¤}("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1±Q"1€±Q"1O’"1üO’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#15#1,5#1��#1˜��#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°X$ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“‰Éä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°X$ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“‰Éä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°X$  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“‰Éä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ›»r’Vv bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Ý€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1tôQÑ.® ‹_Ÿ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßWWËË)?*?4³4³>'?'I™I™S T ^}^}hïhïrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ��  "÷"÷$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¡}!1С}!1?¾!1L?¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1sç!1(sç!1("1¤("1[/"1¬[/"1å8"1(å8"1<"1ô<"1EQ"1€EQ"1ã‘"1üã‘"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1i2#1,i2#1çŒ#1˜çŒ#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1A�&1ÌA�&1S'1S'1¹e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ê¶ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @NŽÛ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ê¶ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @NŽÛ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ê¶  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @NŽÛ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ½Ý•µDd bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÚZ´4ް ð  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜTTÌÌ)D*D4¼4¼>4?4IªIªS T ^–^–h i s‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœŒŒ | |"ô"ô$l%l'ä'ä)\*\,Ô,Ô.L/L1Ä1Ä3<4<6´6´8,9,;¤;¤=>@”@”B C E„E„GüGüItJtLêLêN`O`QÖQÖSLTLVÂVÂX8Y8[®[®]$^$`š`šbce†e†gügüirjrlèlèn^o^qÔqÔsJtJvÀvÀx6y6{¬{¬}"~"€˜€˜‚ƒ…„…„‡ú‡ú‰pŠpŒæŒæŽ\�\‘Ò‘Ò“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1~!1Ð~!1Ÿ¾!1LŸ¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Óç!1(Óç!1q("1¤q("1[/"1¬[/"1å8"1(å8"1<"1ô<"1¥Q"1€¥Q"1C’"1üC’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1[Ú"1´ [Ú"1»þ"10 »þ"1ý+#1Pý+#1i2#1,i2#1§�#1˜§�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£I b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M��ÍÛ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£I b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M��ÍÛ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£I  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M��ÍÛ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~ž½Ý,L bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€èhÐP¸8p° €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜQQÆÆ);*;4°4°>#?#I–I–S T ^|^|hïhïrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒvvëë` `"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?O@OBÂBÂD5E5G¨G¨IJLŽLŽNOQtQtSçSçUZVZXÍXÍZ@[@]³]³_&`&b™b™d e ggiòiòkelenØnØpKqKs¾s¾u1v1x¤x¤z{}Š}Šýý�p‚p„ã„ã†V‡V‰ȉȋ:Œ:ެެ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¡}!1С}!1?¾!1L?¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1sç!1(sç!1("1¤("1[/"1¬[/"1å8"1(å8"1<"1ô<"1EQ"1€EQ"1ã‘"1üã‘"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1=Ú"1´ =Ú"1�þ"10 �þ"1ß+#1Pß+#15#1,5#1}�#1˜}�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†éµ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @NŽÛ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†éµ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @NŽÛ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†éµ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @NŽÛ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ½ÝŸ. bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1šÚ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hèBœövJŠ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÆÆ)<*<4²4²>(?(IœIœST^„^„høhørlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$N%N'Ä'Ä):*:,°,°.&/&1œ1œ346ˆ6ˆ8þ8þ:t;t=ê=ê?`@`BÖBÖDLELGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡Œ‡Œ‰ŠŒtŒtŽèŽè�\‘\“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1~!1Ð~!1¥¾!1L¥¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Óç!1(Óç!1q("1¤q("1[/"1¬[/"1å8"1(å8"1<"1ô<"1¥Q"1€¥Q"1C’"1üC’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#15#1,5#1OŽ#1˜OŽ#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1'€&1Ì'€&1÷Q'1÷Q'1Ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥M b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M��ÍÛ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥M b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M��ÍÛ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥M  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M��ÍÛ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ž¾½ÝXx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¨è€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  ˆpðXØX €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜOOÂÂ)5*54¨4¨>?IŠIŠSûSû]l^lhÝhÝrNsN} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒÿÿrrååX X"Ë"Ë$>%>'±'±)$*$,—,—. / 1}1}3ð3ð5c6c8Ö8Ö:I;I=¼=¼?-@-BžBžDEG€G€IñIñKbLbNÓNÓPDQDSµSµU&V&X—X—Z[]y]y_ê_êa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1§}!1Ч}!1E¾!1LE¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1yç!1(yç!1("1¤("1[/"1¬[/"1å8"1(å8"1<"1ô<"1KQ"1€KQ"1é‘"1üé‘"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1o2#1,o2#1›�#1˜›�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ë· b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @NŽÛ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ë· b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @NŽÛ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ë·  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @NŽÛ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ½ÝšºJj bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1A€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�æfËK°0a¡ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I£I£ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbèbèd]e]gÒgÒiGjGl¼l¼n1o1q¦q¦stv�v�xy{z{z}ï}ïd€d‚قلN…N‡ÇÉ8Š8Œ­Œ­Ž"�"‘—‘—“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1û}!1Ðû}!1™¾!1L™¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Íç!1(Íç!1k("1¤k("1[/"1¬[/"1å8"1(å8"1<"1ô<"1ŸQ"1€ŸQ"1=’"1ü=’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1o2#1,o2#1Ë�#1˜Ë�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1Õ€&1ÌÕ€&1¥R'1¥R'1Me'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜>  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M�‹ËÚ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜>  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M�‹ËÚ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜>   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M�‹ËÚ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ½Ý�¡2R bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€êjÔT¾>�Ð €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áVVËË)@*@4µ4µ>+?+IŸIŸST^‡^‡hûhûrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‹‹ u u"ê"ê$_%_'Ô'Ô)I*I,¾,¾.3/31¨1¨346’6’89;|;|=ñ=ñ?f@fBÛBÛDPEPGÅGÅI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÃbÃd7e7g«g«ijl“l“noq{q{sïsïucvcx×x×zK{K}¿}¿3€3‚§‚§„…‡‘‡‘‰ŠŒ{Œ{ŽðŽð�e‘e“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1§}!1Ч}!1E¾!1LE¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1yç!1(yç!1("1¤("1[/"1¬[/"1å8"1(å8"1<"1ô<"1KQ"1€KQ"1é‘"1üé‘"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#15#1,5#1=Ž#1˜=Ž#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1¥€&1Ì¥€&1uR'1uR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ê¶ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @NŽÛ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ê¶ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @NŽÛ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ê¶  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @NŽÛ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ½Ý€ 1 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¦æ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1•zú_ßDÄF €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÆÆ);*;4°4°>%?%I˜I˜S T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnænæpYqYsÌsÌu?v?x²x²z%{%}˜}˜ € ‚~‚~„ñ„ñ†d‡d‰׉׋JŒJ޽޽�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1~!1Ð~!1Ÿ¾!1LŸ¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1Íç!1(Íç!1k("1¤k("1[/"1¬[/"1å8"1(å8"1<"1ô<"1ŸQ"1€ŸQ"1=’"1ü=’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#15#1,5#1sŽ#1˜sŽ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšB b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M�ŠÊÚ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšB b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M�ŠÊÚ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšB  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M�ŠÊÚ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ½Ý¡ÁZz bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ªê€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¨(’|üfæ:z €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝRRÇÇ)<*<4±4±>&?&I™I™S T ^}^}hïhïrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��wwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD.?.I¢I¢ST^Š^Šhþhþrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²((žž Š Š"#%v%v'ì'ì)b*b,Ø,Ø.N/N1Ä1Ä3:4:6°6°8&9&;œ;œ=>@†@†BúBúDnEnGâGâIVJVLÊLÊN>O>Q²Q²S&T&VšVšXY[‚[‚]ö]ö_j`jbÞbÞdReRgÆgÆi:j:l®l®n"o"q–q–s t v~v~xòxòzf{f}Ú}ÚN€N‚‚„6…6‡ª‡ª‰ŠŒ�Œ�Ž�‘v‘v“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1õ}!1Ðõ}!1“¾!1L“¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Áç!1(Áç!1_("1¤_("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1“Q"1€“Q"11’"1ü1’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1aÚ"1´ aÚ"1Áþ"10 Áþ"1,#1P,#1o2#1,o2#1Ž#1˜Ž#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1�&1Ì�&1ÕR'1ÕR'1}e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�3ÿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M�Ú’Ò bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�3ÿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M�Ú’Ò bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�3ÿ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M�Ú’Ò  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ½Ý|œ(H bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€õuêjß_@€ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii à àWWÎÎ)E*E4¼4¼>5?5I¬I¬S#T#^•^•hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiààWWÎÎ E E ¼ ¼33ªª!!˜˜††ýýt t"ë"ë$b%b'Ù'Ù)P*P,Ç,Ç.>/>1µ1µ3,4,6£6£89;‘;‘=>@@BöBöDmEmGäGäI]J]LÔLÔNKOKQÂQÂS9T9V°V°X'Y'[ž[ž]^`Œ`ŒbcezezgñgñihjhlßlßnQoQqÃqÃs5t5v§v§xy{‹{‹}ý}ýo€o‚á‚á„S…S‡Ňʼn7Š7Œ©Œ©Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1­}!1Э}!1K¾!1LK¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1yç!1(yç!1("1¤("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1KQ"1€KQ"1é‘"1üé‘"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1�Ú"1´ �Ú"1ýþ"10 ýþ"1?,#1P?,#1!5#1,!5#1ýŽ#1˜ýŽ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1u€&1Ìu€&1ER'1ER'1íd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆë· b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @NŽÛ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆë· b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @NŽÛ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆë·  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @NŽÛ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �¡^~5 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1°ð€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1À@°0  �À €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àTTÈÈ)<*<4°4°>"?"I”I”ST^x^xhêhêr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  ˆˆüüp p"ä"ä$X%X'Ì'Ì)@*@,´,´.(/(1œ1œ346„6„8ø8ø:l;l=à=à?R@RBÄBÄD6E6G¨G¨IJLŒLŒNþNþPpQpSâSâUTVTXÆXÆZ8[8]ª]ª_`bŽbŽdegrgriäiäkVlVnÈnÈp:q:s¬s¬uvx�x�z{}t}tææ�X‚X„ʄʆ<‡<‰­‰­‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1õ}!1Ðõ}!1“¾!1L“¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Çç!1(Çç!1e("1¤e("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1™Q"1€™Q"17’"1ü7’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1'5#1,'5#13�#1˜3�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1«€&1Ì«€&1{R'1{R'1#e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�7 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 M�Ú“Ó bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�7 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 M�Ú“Ó bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�7  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 M�Ú“Ó  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ½Ýœ¼Pp bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1µõ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÔTÉI¾>³3õ5 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ü ÜQQÆÆ);*;4°4°>'?'IœIœST^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÜÜQQÆÆ ; ; ° °%%šš„„ùùnnããX X"Í"Í$B%B'·'·),*,,¡,¡./1‹1‹346u6u8ê8ê:_;_=Ô=Ô?I@IB¾B¾D3E3G¨G¨IJL”L”N O Q~Q~SóSóUhVhXÝXÝZR[R]Ç]Ç_<`M?MIÄIÄS9T9^¬^¬his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðyyòòkkää ] ] Ö ÖOOÈÈAAºº33¬¬%% ž ž"#%�%�' ( *‚*‚,û,û.t/t1í1í3f4f6ß6ß8X9X;Ñ;Ñ=J>J@Ã@ÃB@?@I´I´S(T(^œ^œhis„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvððjjää ^ ^ Ø ØRRÌÌFFÀÀ::´´.. ¨ ¨""#"%˜%˜'(*„*„,ú,ú.p/p1æ1æ3\4\6Ò6Ò8H9H;¾;¾=4>4@ª@ªB C E–E–G H J€J€LôLôNhOhQÜQÜSPTPVÄVÄX8Y8[¬[¬] ^ `”`”bce|e|gðgðidjdlØlØnLoLqÀqÀs4t4v¨v¨xy{�{�}~€x€x‚ì‚ì„`…`‡Ô‡Ô‰HŠHŒ¼Œ¼Ž0�0‘¤‘¤“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Å}!1�Å}!1c¾!1c¾!1¹Â!1$¹Â!1‘ç!1�‘ç!1/("1/("1‹,"1$‹,"1cQ"1�cQ"1’"1’"1]–"1$]–"1�Ú"1��Ú"1ýþ"1üýþ"150#1 50#1#1| #1á$1Ü á$1¹ò$1  ¹ò$1óÞ%1¤ óÞ%1á€&1„ á€&1±R'1ȱR'1Ye'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Jd˜ûÇîº b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Jd˜ûÇîº b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Jd˜ûÇîº b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Vv  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww î îeeÜÜ)S*S4Ê4Ê>???I´I´S'T'^˜^˜h i sysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwîîeeÜÜ S S Ê ÊAA¸¸//¦¦””  ‚ ‚"ù"ù$p%p'ç'ç)^*^,Õ,Õ.L/L1Ã1Ã3:4:6±6±8(9(;Ÿ;Ÿ=>@�@�BCE{E{GòGòIgJgLÜLÜNQOQQÆQÆS;T;V°V°X%Y%[š[š]^`„`„b÷b÷djejgÝgÝiPjPlÃlÃn4o4q¥q¥stv‡v‡xøxøzi{i}Ú}ÚK€K‚¼‚¼„-…-‡�‡�‰ŠŒŒŽðŽð�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1ï}!1�ï}!1�¾!1 �¾!1YÅ!1YÅ!1Áç!1€Áç!1_("1ü_("1+/"1+/"1“Q"1p“Q"11’"1ì1’"1ý˜"1ôý˜"1Ù"1`Ù"1eý"1Ìeý"150#1ì50#1#1L #1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�)õ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 à€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�)õ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 à€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1M�)õ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 à €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½Ý  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv î îffÚÚ)N*N4Â4Â>4?4I¦I¦ST^Š^Šhühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvîîffÞÞ V V Î ÎFF¾¾66®®&&žž Ž Ž"#%z%z'î'î)b*b,Ö,Ö.J/J1¾1¾32426¦6¦89;Ž;Ž=>@v@vBêBêD^E^GÒGÒIDJDL¶L¶N(O(QšQšS T V~V~XðXðZb[b]Ô]Ô_F`Fb¸b¸d*e*gœgœijl€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1¿}!1�¿}!1]¾!1 ]¾!1³Â!1³Â!1‘ç!1€‘ç!1/("1ü/("1…,"1…,"1cQ"1pcQ"1’"1ì’"1W–"1ôW–"1‹Ú"1`‹Ú"1ëþ"1Ìëþ"1/0#1ì/0#1û~#1L û~#1ó$1¬ ó$1Ëò$1p Ëò$1ß%1t ß%1Y�&1T Y�&1)S'1˜)S'1Ñe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–ûÇ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–ûÇ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–ûÇ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Hh  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu í íeeÝÝ)U*U4Í4Í>C?CI¹I¹S/T/^¥^¥his‘s‘} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuííeeÝÝ U U Í ÍEE½½55­­%%�� � �"#%}%}'õ'õ)m*m,å,å.]/]1Õ1Õ3M4M6Å6Å8=9=;µ;µ=->-@¥@¥BCE•E•G H JƒJƒLùLùNoOoQåQåS[T[VÑVÑXGYG[½[½]3^3`©`©bce•e•g h j�j�l÷l÷nmomqãqãsYtYvÏvÏxEyE{»{»}1~1€§€§‚ƒ…“…“‡ ˆ ŠŠŒõŒõŽk�k‘á‘á“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1õ}!1�õ}!1“¾!1 “¾!1GÅ!1GÅ!1Çç!1€Çç!1e("1üe("1/"1/"1™Q"1p™Q"17’"1ì7’"1ë˜"1ôë˜"1µÚ"1`µÚ"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ0ü b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1’Ò€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ0ü b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1’Ò€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ0ü b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1’Ò €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1q‘  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo å åZZÏÏ)D*D4·4·>*?*I�I�ST^ƒ^ƒhöhörgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooåå[[ÑÑ G G ½ ½33©©••  ôôi i"Þ"Þ$S%S'È'È)=*=,²,².'/'1œ1œ346†6†8û8û:p;p=ã=ã?V@VBÉBÉD5?5I©I©ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðsséé__ÕÕ K K Á Á77­­##™™……ûûq q"ç"ç$]%]'Ó'Ó)I*I,¿,¿.5/51«1«3!4!6—6—8 9 ;ƒ;ƒ=ù=ù?o@oBåBåD[E[GÑGÑIEJEL¹L¹N-O-Q¡Q¡STV‰V‰XýXýZq[q]å]å_Y`YbÍbÍdAeAgµgµi)j)l�l�noq…q…sùsùumvmxáxázU{U}É}É=€=‚±‚±„%…%‡™‡™‰ Š Œ�Œ�ŽõŽõ�i‘i“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1õ}!1�õ}!1“¾!1 “¾!1GÅ!1GÅ!1Çç!1€Çç!1e("1üe("1/"1/"1™Q"1p™Q"17’"1ì7’"1ë˜"1ôë˜"1µÚ"1`µÚ"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%19€&1T 9€&1 R'1˜ R'1±d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ2þ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1“Ó€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ2þ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1“Ó€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ2þ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1“Ó €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1t”  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ç çZZÍÍ)@*@4±4±>"?"I“I“ST^u^uhæhærUsU} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqçç]]ÓÓ I I ¿ ¿55««!!——  }}ððc c"Ö"Ö$I%I'¼'¼)/*/,¢,¢./1ˆ1ˆ3û3û5n6n8á8á:T;T=Å=Å?6@6B§B§DEG‰G‰IúIúKkLkNÜNÜPMQMS¾S¾U/V/X X Z[]‚]‚_ó_óadbddÕdÕfFgFi·i·k(l(n™n™p q s{s{uìuìw]x]zÎzÎ|?}?°°�!‚!„’„’†‡‰q‰q‹á‹á�QŽQ�Á�Á’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1¹}!1�¹}!1W¾!1 W¾!1¹Â!1¹Â!1‹ç!1€‹ç!1)("1ü)("1‹,"1‹,"1]Q"1p]Q"1û‘"1ìû‘"1]–"1ô]–"1Ú"1`Ú"1ßþ"1Ìßþ"1)0#1ì)0#1õ~#1L õ~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1�&1T �&1çR'1˜çR'1�e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“øÄ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“øÄ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“øÄ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Cc  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq è è__ÖÖ)M*M4Ä4Ä>9?9I®I®S#T#^˜^˜h i s‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqèè__ÖÖ M M Ä Ä;;²²))  ŽŽ | |"ó"ó$j%j'á'á)X*X,Ï,Ï.F/F1½1½34446«6«8"9";™;™=>@‡@‡BþBþDuEuGìGìIaJaLÖLÖNKOKQÀQÀS5T5VªVªXY[”[”] ^ `~`~bóbódhehgÝgÝiRjRlÇlÇn.?.I¢I¢ST^Š^Šhþhþrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnææ^^ÖÖ N N Æ Æ>>¶¶..¦¦�� z z"ï"ï$d%d'Ù'Ù)N*N,Ã,Ã.8/81­1­3"4"6—6—8 9 ;�;�=ö=ö?k@kBàBàDUEUGÊGÊI>J>L²L²N&O&QšQšSTV‚V‚XöXöZj[j]Þ]Þ_R`RbÆbÆd:e:g®g®i"j"l–l–n o q~q~sòsòufvfxÚxÚzN{N}Â}Â6€6‚ª‚ª„…‡’‡’‰ŠŒ|Œ|ŽñŽñ�f‘f“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1³}!1�³}!1Q¾!1 Q¾!1¹Â!1¹Â!1…ç!1€…ç!1#("1ü#("1‹,"1‹,"1WQ"1pWQ"1õ‘"1ìõ‘"1]–"1ô]–"1—Ú"1`—Ú"1÷þ"1Ì÷þ"1)0#1ì)0#1õ~#1L õ~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1±€&1T ±€&1�R'1˜�R'1)e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’õÁ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’õÁ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’õÁ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Vv  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æ\\ÑÑ)F*F4»4»>.?.I¡I¡ST^‡^‡húhúrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  ‚‚øøn n"ä"ä$Y%Y'Î'Î)C*C,¸,¸.-/-1¢1¢346Œ6Œ89;v;v=ë=ë?`@`BÕBÕDJEJG¿G¿I2J2L¥L¥NOQ‹Q‹SþSþUqVqXäXäZW[W]Ê]Ê_=`=b°b°d#e#g–g–i j l|l|nïnïpbqbsÕsÕuHvHx»x»z.{.}¡}¡€‚‡‚‡„ú„ú†m‡m‰à‰à‹SŒSŽÆŽÆ�9‘9“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1û}!1�û}!1™¾!1 ™¾!1AÅ!1AÅ!1Íç!1€Íç!1k("1ük("1/"1/"1ŸQ"1pŸQ"1=’"1ì=’"1å˜"1ôå˜"1ÁÚ"1`ÁÚ"1!ÿ"1Ì!ÿ"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“; b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‘Ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“; b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‘Ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“; b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‘Ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1�¡  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn ä äYYÎÎ)C*C4¸4¸>+?+IžIžST^‚^‚hôhôrfsf} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnääZZÐÐ F F ¼ ¼22¨¨””  ~~óóh h"Ý"Ý$R%R'Ç'Ç)<*<,±,±.&/&1›1›346…6…8ú8ú:o;o=ä=ä?Y@YBÎBÎDCECG¸G¸I+J+LžLžNOQ„Q„S÷S÷UjVjXÝXÝZP[P]Ã]Ã_6`6b¨b¨degŒgŒiþiþkplpnânâpTqTsÆsÆu8v8xªxªz{}Ž}Ž€‚r‚r„ä„ä†V‡V‰ɉɋ<Œ<ޝޝ�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1³}!1�³}!1Q¾!1 Q¾!1³Â!1³Â!1ç!1€ç!1("1ü("1…,"1…,"1QQ"1pQQ"1ï‘"1ìï‘"1W–"1ôW–"1ùØ"1`ùØ"1Yý"1ÌYý"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1í€&1T í€&1½R'1˜½R'1ee'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�õÁ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�õÁ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�õÁ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½Ý  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ë ëaa××)M*M4Á4Á>5?5I©I©ST^‘^‘hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqëëeeßß Y Y Ó ÓMMÇÇAA»»11§§ “ “" # %%'õ'õ)k*k,á,á.W/W1Í1Í3C4C6¹6¹8/9/;¥;¥=>@�@�BCEuEuGéGéI]J]LÑLÑNEOEQ¹Q¹S-T-V¡V¡XY[‰[‰]ý]ý_q`qbåbådYeYgÍgÍiAjAlµlµn)o)q�q�stv…v…xùxùzm{m}á}áU€U‚ɂɄ=…=‡¯‡¯‰"Š"Œ•Œ•Ž�‘{‘{“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1~!1�~!1Ÿ¾!1 Ÿ¾!1AÅ!1AÅ!1Íç!1€Íç!1k("1ük("1/"1/"1ŸQ"1pŸQ"1=’"1ì=’"1å˜"1ôå˜"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœB b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1–Ö€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœB b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1–Ö€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœB b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1–Ö €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1o�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj á áXXÏÏ)F*F4½4½>4?4I«I«S"T"^”^”hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjááXXÏÏ F F ½ ½44««""™™‡‡þþu u"ì"ì$c%c'Ú'Ú)Q*Q,È,È.?/?1¶1¶3-4-6¤6¤89;’;’= > @€@€B÷B÷DnEnGåGåI\J\LÓLÓNJOJQÁQÁS8T8V¯V¯X&Y&[�[�]^`‹`‹bceyeygðgðigjglÞlÞnPoPqÂqÂs4t4v¦v¦xy{Š{Š}ü}ün€n‚à‚à„R…R‡ćĉ6Š6Œ¨Œ¨Ž�‘Œ‘Œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1­}!1�­}!1K¾!1 K¾!1¹Â!1¹Â!1ç!1€ç!1("1ü("1‹,"1‹,"1QQ"1pQQ"1ï‘"1ìï‘"1]–"1ô]–"1ùØ"1`ùØ"1Yý"1ÌYý"1)0#1ì)0#1õ~#1L õ~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1u€&1T u€&1ER'1˜ER'1íd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ò¾ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ò¾ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ò¾ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½Ý  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo ç ç[[ÏÏ)C*C4µ4µ>'?'I™I™S T ^}^}hïhïr_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooçç__×× O O Ç Ç??··//§§�� w w"ë"ë$_%_'Ó'Ó)G*G,»,».///1£1£346‹6‹8ÿ8ÿ:s;s=å=å?W@WBÉBÉD;E;G­G­IJL‘L‘NOQuQuSçSçUYVYXËXËZ=[=]¯]¯_!`!b“b“degwgwiéiék[l[nÍnÍp?q?s±s±u#v#x•x•z{}y}yëë�]‚]„τφ?‡?‰°‰°‹!Œ!Ž’Ž’�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1~!1�~!1Ÿ¾!1 Ÿ¾!1AÅ!1AÅ!1Óç!1€Óç!1q("1üq("1/"1/"1¥Q"1p¥Q"1C’"1ìC’"1å˜"1ôå˜"1ÁÚ"1`ÁÚ"1!ÿ"1Ì!ÿ"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœD b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1—×€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœD b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1—×€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœD b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1—× €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ˆ¨  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ß ßTTÉÉ)>*>4³4³>(?(I�I�ST^„^„höhörhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjßßTTÉÉ > > ³ ³((��‡‡üüqqææ[ ["Ð"Ð$E%E'º'º)/*/,¤,¤./1Ž1Ž346x6x8í8í:b;b=×=×?L@LBÁBÁD6E6G«G«I J L•L•N O QQSôSôUiViXÞXÞZS[S]È]È_=`=b²b²d'e'gœgœijl†l†nønøpjqjsÜsÜuNvNxÀxÀz2{2}¤}¤€‚ˆ‚ˆ„ú„ú†l‡l‰Þ‰Þ‹PŒPŽÂŽÂ�4‘4“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1­}!1�­}!1K¾!1 K¾!1¹Â!1¹Â!1yç!1€yç!1("1ü("1‹,"1‹,"1KQ"1pKQ"1é‘"1ìé‘"1]–"1ô]–"1óØ"1`óØ"1Sý"1ÌSý"1#0#1ì#0#1ï~#1L ï~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%15�&1T 5�&1S'1˜S'1­e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ò¾ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ò¾ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ò¾ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½Ý  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm æ æ__ØØ)Q*Q4Ê4Ê>A?AI¸I¸S-T-^ ^ his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmææ__ØØ Q Q Ê ÊCC¼¼55®®''   ’ ’" # %„%„'ý'ý)v*v,ï,ï.h/h1á1á3Z4Z6Ó6Ó8L9L;Å;Å=>>>@·@·B0C0E©E©G"H"J™J™LMO‡O‡QþQþSuTuVìVìXcYc[Ú[Ú]Q^Q`È`Èb=c=e²e²g'h'jœjœlmo„o„q÷q÷sjtjvÝvÝxPyP{Ã{Ã}6~6€©€©‚ƒ…�…�‡ˆŠrŠrŒäŒäŽV�V‘ȑȓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1~!1�~!1Ÿ¾!1 Ÿ¾!15Å!15Å!1Óç!1€Óç!1q("1üq("1/"1/"1¥Q"1p¥Q"1C’"1ìC’"1Ù˜"1ôÙ˜"1�Ú"1`�Ú"1ýþ"1Ìýþ"1#0#1ì#0#1ï~#1L ï~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥K b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ѐ} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥K b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ѐ} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥K b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ð €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1hˆ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ä ä^^ÔÔ)J*J4À4À>4?4I¨I¨ST^�^�hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjää^^ØØ R R Ì ÌFFÀÀ::´´..¨¨"" œ œ"#%Œ%Œ'(*x*x,î,î.d/d1Ú1Ú3P4P6Æ6Æ8<9<;²;²=(>(@ž@žBCEŠEŠGHJtJtLèLèN\O\QÐQÐSDTDV¸V¸X,Y,[ [ ]^`ˆ`ˆbübüdpepgägäiXjXlÌlÌn@o@q´q´s(t(vœvœxy{„{„}ø}øl€l‚à‚à„T…T‡ȇȉ<Š<Œ°Œ°Ž$�$‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1§}!1�§}!1E¾!1 E¾!1¹Â!1¹Â!1yç!1€yç!1("1ü("1‹,"1‹,"1KQ"1pKQ"1é‘"1ìé‘"1]–"1ô]–"1óØ"1`óØ"1Sý"1ÌSý"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1À&1T À&1“R'1˜“R'1;e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒï» b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒï» b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒï» b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½Ý  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk â âYYÐÐ)G*G4¾4¾>3?3I¨I¨ST^Œ^Œhýhýrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkââYYÐÐ G G ¾ ¾55¬¬##ššˆˆÿÿv v"í"í$d%d'Û'Û)R*R,É,É.@/@1·1·3.4.6¥6¥89;“;“= > @�@�BøBøDoEoGæGæI[J[LÐLÐNEOEQºQºS/T/V¤V¤XY[Ž[Ž]^`x`xbëbëd^e^gÑgÑiDjDl·l·n(o(q™q™s t v{v{xìxìz]{]}Î}Î?€?‚°‚°„!…!‡‘‡‘‰ŠŒsŒsŽäŽä�U‘U“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1~!1�~!1¥¾!1 ¥¾!15Å!15Å!1Óç!1€Óç!1q("1üq("1/"1/"1¥Q"1p¥Q"1C’"1ìC’"1Ù˜"1ôÙ˜"1ÁÚ"1`ÁÚ"1!ÿ"1Ì!ÿ"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥M b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ï€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥M b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ï€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥M b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ï €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1‹«  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âZZÎÎ)B*B4¶4¶>(?(IšIšS T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢’’  ‚ ‚"ú"ú$n%n'â'â)V*V,Ê,Ê.>/>1²1²3&4&6š6š89;‚;‚=ö=ö?j@jBÞBÞDRERGÆGÆI8J8LªLªNOQŽQŽSTVrVrXäXäZV[V]È]È_:`:b¬b¬deg�g�ijltltnænæpXqXsÊsÊuŒ>ްް�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1¡}!1�¡}!1?¾!1 ?¾!1­Â!1­Â!1sç!1€sç!1("1ü("1,"1,"1EQ"1pEQ"1ã‘"1ìã‘"1Q–"1ôQ–"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1A�&1T A�&1S'1˜S'1¹e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠï» b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1G€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠï» b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1G€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠï» b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1G €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½Ý  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áYYÑÑ)I*I4Á4Á>7?7I­I­S#T#^™^™his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‘‘  � �"ù"ù$q%q'é'é)a*a,Ù,Ù.Q/Q1É1É3A4A6¹6¹8191;©;©=!>!@™@™BCE‰E‰GHJwJwLíLíNcOcQÙQÙSOTOVÅVÅX;Y;[±[±]'^'`�`�bce‰e‰gÿgÿiujulëlënaoaq×q×sMtMvÃvÃx9y9{¯{¯}%~%€›€›‚ƒ…‡…‡‡ý‡ý‰sŠsŒéŒéŽ_�_‘Õ‘Õ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1~!1�~!1¥¾!1 ¥¾!15Å!15Å!1Ùç!1€Ùç!1w("1üw("1/"1/"1«Q"1p«Q"1I’"1ìI’"1Ù˜"1ôÙ˜"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®T  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1’Ò€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®T  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1’Ò€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®T  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1’Ò €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½Ý  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84«4«>?I‘I‘ST^w^whêhêr[s[} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=×=×?J@JB½B½D0E0G£G£IJL‰L‰NüNüPoQoSâSâUUVUXÈXÈZ;[;]®]®_!`!b”b”degzgziíiík`l`nÓnÓpFqFs¹s¹u,v,xŸxŸz{}…}…øø�k‚k„Þ„Þ†O‡O‰Á‰Á‹3Œ3ޥޥ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1¡}!1�¡}!1?¾!1 ?¾!1­Â!1­Â!1sç!1€sç!1("1ü("1,"1,"1EQ"1pEQ"1ã‘"1ìã‘"1Q–"1ôQ–"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ì¸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1H€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ì¸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1H€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ì¸ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1H €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½Ý  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝSSÉÉ)?*?4µ4µ>)?)I�I�ST^…^…hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��yyïïe e"Û"Û$Q%Q'Ç'Ç)=*=,³,³.)/)1Ÿ1Ÿ346‹6‹89;w;w=í=í?c@cBÙBÙDOEOGÅGÅI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒuŒuŽéŽé�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1 ~!1� ~!1«¾!1 «¾!15Å!15Å!1Ùç!1€Ùç!1w("1üw("1/"1/"1«Q"1p«Q"1I’"1ìI’"1Ù˜"1ôÙ˜"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1Ÿ$1¬ Ÿ$1wò$1p wò$1±Þ%1t ±Þ%1!€&1T !€&1ñQ'1˜ñQ'1™d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®V" b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1“Ó€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®V" b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1“Ó€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®V" b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1“Ó €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½Ý  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛNNÁÁ)4*44¥4¥>?I‡I‡SøSø]i^ihÚhÚrIsI} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹þþqqääW W"Ê"Ê$=%='°'°)#*#,–,–. / 1|1|3ï3ï5b6b8Õ8Õ:H;H=¹=¹?*@*B›B›D E G}G}IîIîK_L_NÐNÐPAQAS²S²U#V#X”X”Z[]v]v_ç_çaXbXdÉdÉf:g:i«i«kln�n�pþpþrosouàuàwQxQzÂzÂ|3}3¤¤�‚„†„††õ†õˆe‰e‹Õ‹Õ�EŽE�µ�µ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1›}!1�›}!19¾!1 9¾!1­Â!1­Â!1mç!1€mç!1 ("1ü ("1,"1,"1?Q"1p?Q"1Ý‘"1ìÝ‘"1Q–"1ôQ–"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ì¸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ì¸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ì¸ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½Ý  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>-?-I¢I¢ST^Ž^Žhisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbçbçd\e\gÑgÑiFjFl»l»n2o2q§q§stv‘v‘xy{{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1 ~!1� ~!1«¾!1 «¾!1/Å!1/Å!1ßç!1€ßç!1}("1ü}("1/"1/"1±Q"1p±Q"1O’"1ìO’"1Ó˜"1ôÓ˜"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·]) b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Í€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·]) b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Í€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·]) b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Í €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½Ý  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚOOÄÄ)9*94®4®>"?"I–I–ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""šš„„ùùn n"ã"ã$X%X'Í'Í)B*B,·,·.,/,1¡1¡346‹6‹89;u;u=ê=ê?_@_BÔBÔDIEIG¾G¾I2J2L¦L¦NOQŽQŽSTVvVvXêXêZ^[^]Ò]Ò_F`FbÄbÄd8e8g¬g¬i j l”l”noq|q|sðsðudvdxØxØzL{L}À}À4€4‚¨‚¨„…‡�‡�‰ŠŒzŒzŽïŽï�d‘d“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1›}!1�›}!19¾!1 9¾!1­Â!1­Â!1gç!1€gç!1("1ü("1,"1,"19Q"1p9Q"1ב"1ìב"1Q–"1ôQ–"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†éµ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†éµ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†éµ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½Ý  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÅÅ):*:4¯4¯>"?"I•I•ST^^hòhòrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=ß=ß?T@TBÉBÉD>E>G³G³I&J&L™L™N O QQSòSòUeVeXØXØZK[K]¾]¾_1`1b¤b¤degŠgŠiýiýkplpnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1 ~!1� ~!1«¾!1 «¾!1)Å!1)Å!1ßç!1€ßç!1}("1ü}("1û."1û."1±Q"1p±Q"1O’"1ìO’"1͘"1ô͘"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·_+ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ŒÌ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·_+ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ŒÌ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·_+ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ŒÌ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½Ý  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØMMÂÂ)7*74¬4¬>?I’I’S T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?M@MBÂBÂD7E7G¬G¬IJL’L’NOQxQxSëSëU^V^XÑXÑZD[D]·]·_*`*b¤b¤degˆgˆiúiúklllnÞnÞpPqPsÂsÂu4v4x¦x¦z{}Š}Šüü�n‚n„à„à†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1•}!1�•}!13¾!1 3¾!1¡Â!1¡Â!1gç!1€gç!1("1ü("1s,"1s,"19Q"1p9Q"1ב"1ìב"1E–"1ôE–"1ÃØ"1`ÃØ"1#ý"1Ì#ý"10#1ì0#1ã~#1L ã~#1Ã$1¬ Ã$1ëð$1p ëð$1ÉÀ%1t ÉÀ%1á€&1T á€&1×P'1˜×P'1c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„éµ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1D€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„éµ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1D€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„éµ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1D €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½¹  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee ß ßUUÍÍ)C*C4·4·>+?+IŸIŸST^“^“hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯%%›› ‡ ‡"ý"ý$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@ƒ@ƒB÷B÷DkEkGßGßISJSLÇLÇN;O;Q¯Q¯S#T#V—V—X Y [[]ó]ó_g`gbçbçd[e[gÏgÏiCjCl·l·n+o+qŸqŸstv‡v‡xûxûzo{o}ã}ãW€W‚˂˄?…?‡±‡±‰$Š$Œ—Œ—Ž � ‘}‘}“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1~!1�~!1±¾!1 ±¾!1)Å!1)Å!1åç!1€åç!1ƒ("1üƒ("1û."1û."1·Q"1p·Q"1U’"1ìU’"1͘"1ô͘"1Ú"1`Ú"1yþ"1Ôyþ"10#1ô0#1ã~#1T ã~#1Ï$1´ Ï$1÷ð$1x ÷ð$1ÕÀ%1| ÕÀ%1ù€&1\ ù€&1ïP'1 ïP'1—c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀf2 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ï€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀf2 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ï€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀf2 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ï €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1©ñ[  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ Õ ÕLLÍÍ)D*D4»4»>2?2I©I©S T ^ž^žhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÕÕLLÃà : : ± ±((ŸŸ��{{òòi i"à"à$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÑLÑNHOHQ¿Q¿S6T6V­V­X$Y$[›[›]^`‰`‰bcewewgîgîiejelÜlÜnZoZqÌqÌs>t>v°v°x"y"{”{”}~€x€x‚ê‚ê„\…\‡·Ή@Š@Œ²Œ²Ž$�$‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1•}!1�•}!13¾!1 3¾!1¡Â!1¡Â!1aç!1€aç!1ÿ'"1üÿ'"1s,"1s,"13Q"1p3Q"1Ñ‘"1ìÑ‘"1E–"1ôE–"1ÃØ"1`ÃØ"1#ý"1Ì#ý"10#1ì0#1ã~#1L ã~#1±$1¬ ±$1Ùð$1p Ùð$1·À%1t ·À%1i€&1T i€&1_P'1˜_P'1c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒæ² b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1E€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒæ² b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1E€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒæ² b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1E €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1½¹  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛOOÇÇ);*;4­4­>?I‘I‘ST^�^�hóhórcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››ƒƒ÷÷k k"ß"ß$W%W'Ë'Ë)?*?,³,³.'/'1›1›346ƒ6ƒ8÷8÷:k;k=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b—b—d e g{g{iíiík_l_nÑnÑpCqCsµsµu'v'x™x™z { }}}}ïï�a‚a„Ó„Ó†C‡C‰´‰´‹%Œ%Ž–Ž–�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1~!1�~!1±¾!1 ±¾!1)Å!1)Å!1åç!1€åç!1ƒ("1üƒ("1û."1û."1·Q"1p·Q"1U’"1ìU’"1͘"1ô͘"1OÚ"1`OÚ"1¯þ"1Ô¯þ"10#1ô0#1ã~#1T ã~#1½$1´ ½$1åð$1x åð$1ÃÀ%1| ÃÀ%1Ÿ€&1\ Ÿ€&1•P'1 •P'1=c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀh4 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ѐ} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀh4 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ѐ} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀh4 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ð €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1À“G  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ Ó ÓHHÅÅ):*:4¯4¯>$?$I™I™ST^Œ^Œhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÓÓHH½½ 2 2 § §‘‘{{ððeeÚÚO O"Ä"Ä$A%A'¶'¶)+*+, , ./1Š1Š3ÿ3ÿ5t6t8é8é:^;^=Ó=Ó?H@HB½B½D2E2G§G§IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚noqrqrsäsäuVvVxÈxÈz:{:}¬}¬€‚�‚�„…‡t‡t‰æ‰æ‹XŒXŽÊŽÊ�<‘<“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„+i!1Y}!1Y}!1÷½!1|÷½!1[Ã!1„[Ã!1‘Æ!1ð‘Æ!1Ð!1|Ð!1QÓ!1HQÓ!1%ç!1Ô%ç!1Ã'"1PÃ'"1--"1X--"1c0"1Äc0"1í9"1Pí9"1#="1#="1÷P"1¨÷P"1•‘"1$ •‘"1ÿ–"1, ÿ–"15š"1˜ 5š"1¿£"1$ ¿£"1õ¦"1ð õ¦"17Ô"1| 7Ô"1gÚ"1XgÚ"1Çþ"1ÔÇþ"1 ,#1ô ,#1a:#1Ða:#1o#1\o#1OŽ#1€OŽ#1M™#1¬M™#1Ã$18Ã$1›ò$1ü›ò$1ÕÞ%1ÕÞ%1Oì%1àOì%1Ï€&1lÏ€&1ŸR'1°ŸR'1Ge'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P·ƒ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¿ÿ bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P·ƒ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¿ÿ  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P·ƒ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¿ÿ bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 a�òBb bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Q‘â"ø8uµH  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dQÑ"¢ósÄD•fæ7·ˆÒR‹ Û[DÄ®.ý}� T Ô ' § ú z Í M   Ž Î q ± B•Õ(h bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ ¥ ¥LLËË)@*@4µ4µ>+?+IŸIŸST^“^“his}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__××OOÇÇ ? ? … …ýýuuííee  ��öök k"à"à$_%_'Ô'Ô)I*I,¾,¾.3/31¨1¨346’6’89;|;|=ñ=ñ?f@fBÛBÛDPEPGÅGÅI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÏbÏdCeCg·g·i+j+lŸlŸnoq‡q‡sûsûuovoxãxãzW{W}Ë}Ë?€?‚³‚³„'…'‡�‡�‰ŠŒ‡Œ‡ŽüŽü�q‘q“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1+~!1è+~!1ɾ!1dɾ!1wÅ!1lwÅ!1Ï!1èÏ!17Ò!1´7Ò!1÷ç!1@÷ç!1•("1¼•("1I/"1ÄI/"1Ó8"1@Ó8"1 <"1  <"1ÉQ"1˜ÉQ"1g’"1g’"1™"1 ™"1¥¢"1˜ ¥¢"1Û¥"1d Û¥"1Ó"1ð Ó"1aÚ"1Ì aÚ"1Áþ"1H Áþ"1,#1h,#1U:#1DU:#1o#1Ðo#1Ž#1ôŽ#1™#1 ™#1É$1¬É$1¡ò$1p¡ò$1ÛÞ%1tÛÞ%1Uì%1TUì%1ÿ€&1àÿ€&1ÏR'1$ÏR'1we'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¨Ü„P b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O��ÁÞ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¨Ü„P b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O��ÁÞ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¨Ü„P  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O��ÁÞ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¿ßŠªDd bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1O�Þm­:zü<  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dOÏžím¼<‹ ZÚ)©øxÇG”–cãeå2²4 ´  �  ƒ Ò R î . ¡ ! p ð ?  Î ]�í- bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÉÉ)*4·4·>.?.I£I£ST^™^™hisƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$]%]'Ô'Ô)K*K,Â,Â.9/91~1~3õ3õ5l6l8ã8ã:Z;Z=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbèbèd]e]gÒgÒiGjGl¼l¼n=o=q²q²s't'vœvœxy{†{†}û}ûp€p‚å‚å„Z…Z‡χωDŠDŒ¹Œ¹Ž.�.‘£‘£“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„+i!1S}!1S}!1ñ½!1|ñ½!1[Ã!1„[Ã!1‘Æ!1ð‘Æ!1Ð!1|Ð!1QÓ!1HQÓ!1%ç!1Ô%ç!1Ã'"1PÃ'"1--"1X--"1c0"1Äc0"1í9"1Pí9"1#="1#="1÷P"1¨÷P"1•‘"1$ •‘"1ÿ–"1, ÿ–"15š"1˜ 5š"1¿£"1$ ¿£"1õ¦"1ð õ¦"17Ô"1| 7Ô"1IÚ"1XIÚ"1©þ"1Ô©þ"1ë+#1ôë+#1ý:#1Ðý:#1¹o#1\¹o#1[‘#1€[‘#1Yœ#1¬Yœ#1±$18±$1‰ò$1ü‰ò$1ÃÞ%1ÃÞ%1=ì%1à=ì%1‡€&1l‡€&1WR'1°WR'1ÿd'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O´€ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¿ÿ bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O´€ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¿ÿ  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O´€  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¿ÿ bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �¡0(H bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1zº4tì,q± ¤ä  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dzútônîhèbâ\ÜVÖPÐHȶ6 @ À o ï 8 ¸ ( ¨ 0 ° ã c ( ¨     G ˜� JÄ|¼4t bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ ¢ ¢JJÇÇ)<*<4±4±>&?&I™I™ST^‰^‰hûhûroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 z zððffÜÜRRúúooääY Y"Î"Î$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD%?%I˜I˜S T ^Š^Šhýhýrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$N%N'Ã'Ã)8*8,­,­."/"1d1d3Ù3Ù5N6N8Ã8Ã:8;8=à=à?U@UBÊBÊD?E?G´G´I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnònòpeqesØsØuKvKx¾x¾z1{1}¤}¤€‚ЂЄý„ý†p‡p‰ã‰ã‹VŒVŽÉŽÉ�<‘<“& b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 ' bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1•}!1è•}!13¾!1d3¾!1}Å!1l}Å!1Ï!1èÏ!1=Ò!1´=Ò!1aç!1@aç!1ÿ'"1¼ÿ'"1O/"1ÄO/"1Ù8"1@Ù8"1<"1 <"13Q"1˜3Q"1Ñ‘"1Ñ‘"1!™"1 !™"1«¢"1˜ «¢"1á¥"1d á¥"1#Ó"1ð #Ó"1…Ú"1Ì …Ú"1åþ"1H åþ"1',#1h',#1[:#1D[:#1o#1Ðo#1Ž#1ôŽ#1™#1 ™#1Ã$1¬Ã$1›ò$1p›ò$1ÕÞ%1tÕÞ%1Oì%1TOì%1í€&1àí€&1½R'1$½R'1ee'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒê¶ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 PO�Þ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒê¶ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 PO�Þ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒê¶  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 PO�Þ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 f†ü`€ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1O�Þÿ?m­ü<  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dOÏžím¼<‹ ZÚ)©øxáaÇGœ–W×’eå4 ´  ƒ Ò R ¡ ! p ð ¯ ï ?  Î ]�ì, bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Ö ÖMMÄÄ);*;4²4²>+?+IpIpST^•^•hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÖÖMMÄÄ ; ; ² ²))  ŽŽ||óój j"á"á$X%X'Ï'Ï)F*F,½,½.4/41«1«3"4"6™6™89;‡;‡=þ=þ?u@uBìBìDcEcGÚGÚISJSLÊLÊNAOAQ¸Q¸S/T/VtVtXëXëZb[b]Ù]Ù_P`Pbùbùdpepgçgçi^j^lÕlÕnQoQqÃqÃs5t5v§v§xy{‹{‹}ý}ýo€o‚á‚á„S…S‡Ňʼn7Š7Œ©Œ©Ž�‘�‘�“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1%~!1è%~!1þ!1dþ!1}Å!1l}Å!1Ï!1èÏ!1=Ò!1´=Ò!1ñç!1@ñç!1�("1¼�("1O/"1ÄO/"1Ù8"1@Ù8"1<"1 <"1ÃQ"1˜ÃQ"1a’"1a’"1!™"1 !™"1«¢"1˜ «¢"1á¥"1d á¥"1#Ó"1ð #Ó"1‹Ú"1Ì ‹Ú"1ëþ"1H ëþ"1-,#1h-,#1Ù:#1DÙ:#1•o#1Еo#1ƒ�#1ôƒ�#1�›#1 �›#1½$1¬½$1•ò$1p•ò$1ÏÞ%1tÏÞ%1Iì%1TIì%1Õ€&1àÕ€&1¥R'1$¥R'1Me'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�ÑyE b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O�¡áÞ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�ÑyE b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O�¡áÞ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�ÑyE  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O�¡áÞ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �­a�Jj bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1m­Z*jÉ x¸  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dmíZÚGÇ4´!¡Žû{w÷èhDÄ×W‘ÆF Þ^ µ 5 ¤ $ “  ‚  q ñ ù 9`àM�ú:©éX˜ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff à àVVÌÌ)B*B4¸4¸>,?,I I ST^X^Xhþhþrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffààZZÔÔ N N È ÈBB¼¼66°°&&œœ ˆ ˆ"þ"þ$t%t'ê'ê)`*`,Ö,Ö.L/L1Â1Â38486®6®8$9$;š;š=>@„@„BøBøDlElGàGàITJTLÈLÈN?IaIaS T ^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÔÔII¾¾ 3 3 ¨ ¨’’||ññffÛÛP P"Å"Å$:%:'¯'¯)$*$,™,™./1ƒ1ƒ3ø3ø5m6m8â8â:W;W=Ì=Ì?A@AB¶B¶D+E+G G IJLŒLŒNOQvQvSëSëU-V-X¢X¢Z[]Œ]Œ_`b©b©deg“g“ijl}l}n÷n÷piqisÛsÛuMvMx¿x¿z1{1}£}£€‚‡‚‡„ù„ù†k‡k‰݉݋OŒOŽÁŽÁ�3‘3“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1%~!1è%~!1þ!1dþ!1}Å!1l}Å!1Ï!1èÏ!1=Ò!1´=Ò!1÷ç!1@÷ç!1•("1¼•("1O/"1ÄO/"1Ù8"1@Ù8"1<"1 <"1ÉQ"1˜ÉQ"1g’"1g’"1!™"1 !™"1«¢"1˜ «¢"1á¥"1d á¥"1#Ó"1ð #Ó"1[Ú"1Ì [Ú"1»þ"1H »þ"1ý+#1hý+#1y:#1Dy:#15o#1Ð5o#1µŽ#1ôµŽ#1³™#1 ³™#1¥$1¬¥$1}ò$1p}ò$1·Þ%1t·Þ%11ì%1T1ì%1]€&1à]€&1-R'1$-R'1Õd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŸÓ}I b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O� àÞ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŸÓ}I b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O� àÞ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŸÓ}I  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O� àÞ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‘±6V2R bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1X˜ð0†Æ·÷ \  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dXØ0°ˆà`¸8�hè@À–ìlÂB ˜  « + n î x ø D Ä E Å  š  ’ ð p È ` †Æö6ŒÌ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜPPÄÄ)8*84¬4¬>?I�I�ST^E^Ehêhêr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœ„„øøl l"à"à$T%T'È'È)<*<,°,°.$/$1˜1˜3 4 6€6€8ô8ô:h;h=Ü=Ü?N@NBÀBÀD2E2G¤G¤IJLˆLˆNúNúPlQlSÞSÞUPVPXÂXÂZ4[4]¦]¦_`bŽbŽdegrgriäiäkVlVn•n•pqsysyuëuëw]x]z{}t}tææ�X‚X„ʄʆ<‡<‰­‰­‹ŒŽ�Ž��‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1›}!1Л}!19¾!1L9¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1gç!1(gç!1("1¤("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"19Q"1€9Q"1ב"1üב"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1�Ú"1´ �Ú"1ýþ"10 ýþ"1?,#1P?,#1[7#1,[7#1÷‹#1˜÷‹#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1ù€&1Ìù€&1ÉR'1ÉR'1qe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„ë· b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TO�Þ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„ë· b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TO�Þ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„ë·  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TO�Þ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 k‹&o� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ü<€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1êj§'dä!¡Ì  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ý ÝWWÍÍ)C*C4¹4¹>/?/I£I£ST^‹^‹hÿhÿrsss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÝÝWWÑÑ K K Å Å??¹¹33­­''¡¡ • •"#%…%…'û'û)q*q,ç,ç.]/]1Ó1Ó3I4I6¿6¿8595;«;«=!>!@—@—B C EƒEƒGùGùIoJoLãLãNWOWQËQËS?T?V³V³X'Y'[›[›]^`ƒ`ƒb÷b÷dkekgßgßiSjSlÇlÇn;o;q¯q¯s#t#v—v—x y {{}ó}óg€g‚Û‚Û„O…O‡ÇÉ7Š7Œ«Œ«Ž�‘“‘““ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1~!1Ð~!1·¾!1L·¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1ëç!1(ëç!1‰("1¤‰("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1½Q"1€½Q"1[’"1ü[’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1©Ú"1´ ©Ú"1 ÿ"10  ÿ"1K,#1PK,#1y7#1,y7#1óŒ#1˜óŒ#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1)�&1Ì)�&1ùR'1ùR'1¡e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Æn: b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O��Íâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Æn: b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O��Íâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Æn:  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O��Íâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �°Pp€  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1X€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ZÚ%¥ðp»; È  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff ß ßXXÑÑ)J*J4Ã4Ã>7@°@°B)C)E¢E¢GHJ”J”L M O‚O‚QùQùSpTpVçVçX^Y^[Õ[Õ]L^L`Ã`Ãb8c8e­e­g"h"j—j—l m ooqòqòsetevØvØxKyK{¾{¾}1~1€¤€¤‚ƒ…ЅЇý‡ý‰oŠoŒáŒáŽS�S‘őœ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1•}!1Е}!13¾!1L3¾!1}Å!1T}Å!1Ï!1ÐÏ!1=Ò!1œ=Ò!1gç!1(gç!1("1¤("1O/"1¬O/"1Ù8"1(Ù8"1<"1ô<"19Q"1€9Q"1ב"1üב"1!™"1 !™"1«¢"1€ «¢"1á¥"1L á¥"1#Ó"1Ø #Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1×9#1,×9#1uŒ#1˜uŒ#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„éµ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UO�Þ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„éµ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UO�Þ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„éµ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UO�Þ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‹«Tt4T bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1y¹ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Þ^ ›  X Ø  • I ‰  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSÇÇ);*;4¯4¯>#?#I•I•ST^y^yhëhër]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$g%g'Û'Û)O*O,Ã,Ã.7/71«1«346“6“89;{;{=ï=ï?c@cB×B×DKEKG¿G¿I3J3L¥L¥NOQ‰Q‰SûSûUmVmXßXßZQ[Q]Ã]Ã_5`5b§b§deg‹g‹iýiýkolonánápSqSsÅsÅu7v7x©x©z{}�}�ÿÿ�q‚q„ã„ã†U‡U‰ljNj9Œ9ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1~!1Ð~!1½¾!1L½¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1ñç!1(ñç!1�("1¤�("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1ÃQ"1€ÃQ"1a’"1üa’"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#1:#1,:#1}�#1˜}�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1·€&1Ì·€&1‡R'1‡R'1/e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”Èr> b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O�ŽÎã# bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”Èr> b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O�ŽÎã# bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”Èr>  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O�ŽÎã#  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �°Yy0P bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1£ã €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1†  Q Ñ  œ ç g S“ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I£I£ST^‡^‡høhørjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbæbædYeYgÌgÌi?j?l²l²n#o#q”q”stvvvvxçxçzX{X}É}É:€:‚«‚«„…‡އމÿ‰ÿ‹pŒpŽáŽá�R‘R“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1›}!1Л}!19¾!1L9¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1mç!1(mç!1 ("1¤ ("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1?Q"1€?Q"1Ý‘"1üÝ‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1=Ú"1´ =Ú"1�þ"10 �þ"1ß+#1Pß+#1m7#1,m7#1‡Œ#1˜‡Œ#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ì¸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 EQ‘à  bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ì¸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 EQ‘à  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ì¸  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 EQ‘à   bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¿ßr’2 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1H€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  æf¬,ròX ˜  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ö ÖKKÀÀ)5*54ª4ª>?I�I�ST^v^vhéhér\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††ûûppååZ Z"Ï"Ï$D%D'¹'¹).*.,£,£./1�1�346w6w8ì8ì:a;a=Ö=Ö?I@IB¼B¼D/E/G¢G¢IJLˆLˆNûNûPnQnSáSáUTVTXÇXÇZ:[:]­]­_ ` b“b“degygyiìiìk_l_nÒnÒpEqEs¸s¸u+v+xžxžz{}„}„÷÷�j‚j„݄݆P‡P‰‰‹4Œ4ަަ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1~!1Ð~!1±¾!1L±¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1åç!1(åç!1ƒ("1¤ƒ("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1·Q"1€·Q"1U’"1üU’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1µÚ"1´ µÚ"1ÿ"10 ÿ"1W,#1PW,#1‘7#1,‘7#1­�#1˜­�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1;�&1Ì;�&1 S'1 S'1³e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»c/ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 P�–Öï/ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»c/ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 P�–Öï/ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»c/  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 P�–Öï/  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 “³Rr‡§ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1,l€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1°0„XØ, ¬ |¼ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÎÎ)F*F4¾4¾>6?6I¬I¬S"T"^˜^˜his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$n%n'æ'æ)^*^,Ö,Ö.N/N1Æ1Æ3>4>6¶6¶8.9.;¦;¦=>@–@–BCE†E†GþGþIvJvLìLìNbObQØQØSNTNVÄVÄX:Y:[°[°]&^&`œ`œbceˆeˆgþgþitjtlêlên`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1›}!1Л}!19¾!1L9¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1mç!1(mç!1 ("1¤ ("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1?Q"1€?Q"1Ý‘"1üÝ‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1ï9#1,ï9#1 �#1˜ �#1™$1$™$1qò$1èqò$1«Þ%1ì«Þ%1€&1Ì€&1ëQ'1ëQ'1“d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ê¶ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 DP�ß bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ê¶ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 DP�ß bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ê¶  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 DP�ß  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¿ß�°:Z bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ŽÎ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18 ¸ þ ~ Ä D Š Þ  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØKK¾¾)1*14¤4¤>?I†I†S÷S÷]h^hhÙhÙrJsJ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆûûnnááT T"Ç"Ç$:%:'­'­) * ,“,“./1y1y3ì3ì5_6_8Ò8Ò:E;E=¸=¸?)@)BšBšD E G|G|IíIíK^L^NÏNÏP@Q@S±S±U"V"X“X“Z[]u]u_æ_æaWbWdÈdÈf9g9iªiªklnŒnŒpýpýrnsnußußwPxPzÁzÁ|2}2££�‚„…„…†ö†öˆf‰f‹Ö‹Ö�FŽF�¶�¶’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1~!1Ð~!1·¾!1L·¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ëç!1(ëç!1‰("1¤‰("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1½Q"1€½Q"1[’"1ü[’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#1%:#1,%:#1=Ž#1˜=Ž#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½g3 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘•Õð0 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½g3 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘•Õð0 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½g3  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘•Õð0  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �¯.Nv– bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1À €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 € Ô T ¨ ( | ü P €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÆÆ)<*<4²4²>(?(IœIœST^„^„høhørlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$N%N'Ä'Ä):*:,°,°.&/&1œ1œ346ˆ6ˆ8þ8þ:t;t=ê=ê?`@`BÖBÖDLELGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡Œ‡Œ‰ŠŒtŒtŽèŽè�\‘\“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¡}!1С}!1?¾!1L?¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1sç!1(sç!1("1¤("1y/"1¬y/"19"1(9"19<"1ô9<"1EQ"1€EQ"1ã‘"1üã‘"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#1…7#1,…7#1S�#1˜S�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†í¹ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Ig§@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†í¹ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Ig§@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†í¹  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Ig§@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¿ßu•: bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1^€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1xøIÉšëk ^ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛPPÅÅ):*:4¯4¯>%?%I™I™S T ^�^�hõhõrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››……úúo o"ä"ä$Y%Y'Î'Î)C*C,¸,¸.-/-1¢1¢346Œ6Œ89;v;v=ë=ë?`@`BÕBÕDJEJG¿G¿I5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`Ib½b½d1e1g¥g¥ijl�l�noququséséu]v]xÑxÑzE{E}¹}¹-€-‚¡‚¡„…‡‹‡‹‰ŠŒuŒuŽêŽê�_‘_“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1 ~!1Ð ~!1«¾!1L«¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1ßç!1(ßç!1}("1¤}("1[/"1¬[/"1å8"1(å8"1<"1ô<"1±Q"1€±Q"1O’"1üO’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1µÚ"1´ µÚ"1ÿ"10 ÿ"1W,#1PW,#1©7#1,©7#1sŽ#1˜sŽ#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1�&1Ì�&1ÕR'1ÕR'1}e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°X$ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 P�‡Çâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°X$ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 P�‡Çâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°X$  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 P�‡Çâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 –¶Tt‚¢ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1B‚€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ˆçgÆF ¥ % B‚ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ý ÝTTËË)B*B4¹4¹>0?0I¥I¥ST^�^�hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÝÝTTËË B B ¹ ¹00§§••  ƒƒúúq q"è"è$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubêbêd_e_gÔgÔiIjIl¾l¾n3o3q¨q¨stv’v’xy{|{|}ñ}ñf€f‚Û‚Û„P…P‡Ňʼn:Š:Œ¯Œ¯Ž$�$‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¡}!1С}!1?¾!1L?¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1sç!1(sç!1("1¤("1y/"1¬y/"19"1(9"19<"1ô9<"1EQ"1€EQ"1ã‘"1üã‘"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1mÚ"1´ mÚ"1Íþ"10 Íþ"1,#1P,#1:#1,:#1ã�#1˜ã�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ë· b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jf¦þ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ë· b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jf¦þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ë·  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jf¦þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¿ß•µ@` bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¯ï €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¼ < � ^ Þ / ¯ ¯ï €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84­4­>"?"I•I•ST^y^yhëhër_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?N@NBÃBÃD8E8G­G­I"J"L•L•NOQ{Q{SîSîUaVaXÔXÔZG[G]º]º_-`-bŸbŸdegƒgƒiõiõkglgnÙnÙpKqKs½s½u/v/x¡x¡z{}…}…÷÷�i‚i„Û„Û†O‡O‰‰‹5Œ5ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1~!1Ð~!1±¾!1L±¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1ßç!1(ßç!1}("1¤}("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1±Q"1€±Q"1O’"1üO’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1©Ú"1´ ©Ú"1 ÿ"10  ÿ"1K,#1PK,#1I:#1,I:#1�#1˜�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²\( b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘‡Çä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²\( b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘‡Çä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²\(  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘‡Çä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ž®,Lu• bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1á! €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1„  c ã B  ! ¡ á! €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÆÆ);*;4°4°>%?%I˜I˜S T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnænæpYqYsÌsÌu?v?x²x²z%{%}˜}˜ € ‚~‚~„ñ„ñ†d‡d‰׉׋JŒJ޽޽�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1§}!1Ч}!1E¾!1LE¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1yç!1(yç!1("1¤("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1KQ"1€KQ"1é‘"1üé‘"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1�7#1,�7#1Ž#1˜Ž#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ù€&1Ìù€&1ÉR'1ÉR'1qe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡îº b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @P�ß bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡îº b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @P�ß bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡îº  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @P�ß  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¿ßzš$D bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations14t€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÐP¬,ˆ d ä ä$ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>/?/I¦I¦ST^�^�hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIWJWLÎLÎNEOEQ¼Q¼S3T3VªVªX!Y![˜[˜]^`†`†býbýdtetgëgëibjblÙlÙnKoKq½q½s/t/v¡v¡xy{…{…}÷}÷i€i‚Û‚Û„M…M‡¿‡¿‰1Š1Œ£Œ£Ž�‘‡‘‡“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1~!1Ð~!1¥¾!1L¥¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Ùç!1(Ùç!1w("1¤w("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1«Q"1€«Q"1I’"1üI’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1Á7#1,Á7#1?�#1˜?�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1ç€&1Ìç€&1·R'1·R'1_e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥M b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O��Ïß bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥M b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O��Ïß bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥M  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O��Ïß  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ™¹¿ßVv bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1X˜€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`àJÊ4 ´  ž H €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ä äZZÐÐ)F*F4¼4¼>0?0I¤I¤ST^Œ^Œhistst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjää^^ØØ R R Ì ÌFFÀÀ::´´**   Œ Œ"#%x%x'î'î)d*d,Ú,Ú.P/P1Æ1Æ3<4<6²6²8(9(;ž;ž=>@ˆ@ˆBüBüDpEpGäGäIXJXLÌLÌN@O@Q´Q´S(T(VœVœXY[„[„]ø]ø_l`lbàbàdTeTgÈgÈi#?#I˜I˜S T ^^hñhñrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccØØMM 7 7 ¬ ¬!!––  €€õõjjßßT T"É"É$>%>'³'³)(*(,�,�./1‡1‡3ü3ü5q6q8æ8æ:[;[=Ð=Ð?E@EBºBºD/E/G¤G¤IJL�L�NOQzQzSïSïUdVdXÙXÙZN[N]Ã]Ã_8`8b­b­d"e"g—g—i j l�l�nónópeqes×s×uIvIx»x»z-{-}Ÿ}Ÿ€‚ƒ‚ƒ„õ„õ†g‡g‰ىًKŒK޽޽�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1 ~!1Ð ~!1«¾!1L«¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Ùç!1(Ùç!1w("1¤w("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1«Q"1€«Q"1I’"1üI’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1[Ú"1´ [Ú"1»þ"10 »þ"1ý+#1Pý+#1m:#1,m:#1ç�#1˜ç�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1o€&1Ìo€&1?R'1?R'1çd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§Q b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O��Ïß bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§Q b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O��Ïß bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§Q  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O��Ïß  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �­¿ß*J bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1B €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ˆ ò r Ü \ Æ F²ò €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àTTÈÈ)<*<4°4°>"?"I”I”ST^x^xhêhêr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  ˆˆüüp p"ä"ä$X%X'Ì'Ì)@*@,´,´.(/(1œ1œ346„6„8ø8ø:l;l=à=à?R@RBÄBÄD6E6G¨G¨IJLŒLŒNþNþPpQpSâSâUTVTXÆXÆZ8[8]ª]ª_`bŽbŽdegrgriäiäkVlVnÈnÈp:q:s¬s¬uvx�x�z{}t}tææ�X‚X„ʄʆ<‡<‰­‰­‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1­}!1Э}!1K¾!1LK¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1ç!1(ç!1("1¤("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1QQ"1€QQ"1ï‘"1üï‘"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1»7#1,»7#1åŽ#1˜åŽ#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆï» b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @P�ß bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆï» b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @P�ß bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆï»  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @P�ß  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¿ßŸ.N bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1K‹€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1)©�÷w Þ ^ «ë €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á á[[ÑÑ)G*G4½4½>3?3I§I§ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±++¥¥ ™ ™"#%‰%‰'ÿ'ÿ)u*u,ë,ë.a/a1×1×3M4M6Ã6Ã8999;¯;¯=%>%@›@›BCE‡E‡GýGýIsJsLçLçN[O[QÏQÏSCTCV·V·X+Y+[Ÿ[Ÿ]^`‡`‡bûbûdoeogãgãiWjWlËlËn?o?q³q³s't'v›v›xy{ƒ{ƒ}÷}÷k€k‚߂߄S…S‡LJlj;Š;Œ¯Œ¯Ž#�#‘—‘—“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1~!1Ð~!1Ÿ¾!1LŸ¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Íç!1(Íç!1k("1¤k("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1ŸQ"1€ŸQ"1=’"1ü=’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#1ß7#1,ß7#1 �#1˜ �#1á$1$á$1¹ò$1è¹ò$1óÞ%1ìóÞ%1;�&1Ì;�&1 S'1 S'1³e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšB b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O�‰ÉÞ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšB b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O�‰ÉÞ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšB  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O�‰ÉÞ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¿ß”´Ll bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1n®€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¸8­- ¢ " —  Î €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ã ã\\ÕÕ)N*N4Ç4Ç>@?@I·I·S,T,^Ÿ^Ÿhis…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjãã\\ÕÕ N N Ç Ç@@¹¹22««$$�� � �"#%�%�'ú'ú)s*s,ì,ì.e/e1Þ1Þ3W4W6Ð6Ð8I9I;Â;Â=;>;@´@´B-C-E¦E¦GHJ˜J˜LMO†O†QýQýStTtVëVëXbYb[Ù[Ù]P^P`Ç`Çb'?'I™I™S T ^}^}hïhïrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ��  "÷"÷$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1~!1Ð~!1Ÿ¾!1LŸ¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Óç!1(Óç!1q("1¤q("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1¥Q"1€¥Q"1C’"1üC’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1‘:#1,‘:#1¿�#1˜¿�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœF b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O�ˆÈÞ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœF b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O�ˆÈÞ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœF  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O�ˆÈÞ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¿ßˆ¨ @ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1#c €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Œ �  v ö këƒÃ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>2?2I§I§ST^‹^‹hühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbêbêd]e]gÐgÐiCjCl¶l¶n'o'q˜q˜s t vzvzxëxëz\{\}Í}Í>€>‚¯‚¯„ … ‡’‡’‰ŠŒtŒtŽåŽå�V‘V“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1³}!1г}!1Q¾!1LQ¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ç!1(ç!1("1¤("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1QQ"1€QQ"1ï‘"1üï‘"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#1Ó7#1,Ó7#1±�#1˜±�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ð¼ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @P�ß bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ð¼ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @P�ß bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ð¼  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @P�ß  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‚¢4Tb‚ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1` €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ròd ä V Ö p° €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94®4®>!?!I”I”ST^z^zhíhír`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?M@MBÀBÀD3E3G¦G¦IJLŒLŒNÿNÿPrQrSåSåUXVXXËXËZ>[>]±]±_$`$b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†T‡T‰ƉƋ8Œ8ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1û}!1Ðû}!1™¾!1L™¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1Çç!1(Çç!1e("1¤e("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1™Q"1€™Q"17’"1ü7’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1÷7#1,÷7#1Ñ�#1˜Ñ�#1ç$1$ç$1¿ò$1è¿ò$1ùÞ%1ìùÞ%1M�&1ÌM�&1S'1S'1Åe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�7 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O�Þ‘Ñ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�7 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O�Þ‘Ñ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�7  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O�Þ‘Ñ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¿ß‹«6V bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1„Ä€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1� �  �  � ”Ô €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âZZÒÒ)J*J4Â4Â>:?:I°I°S&T&^œ^œhisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢’’  ‚ ‚"ú"ú$r%r'ê'ê)b*b,Ú,Ú.R/R1Ê1Ê3B4B6º6º8292;ª;ª=">"@š@šBCEŠEŠGHJzJzLðLðNfOfQÜQÜSRTRVÈVÈX>Y>[´[´]*^*` ` bceŒeŒghjxjxlîlîndodqÚqÚsPtPvÆvÆx?IŠIŠSûSû]l^lhÝhÝrNsN} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒÿÿrrååX X"Ë"Ë$>%>'±'±)$*$,—,—. / 1}1}3ð3ð5c6c8Ö8Ö:I;I=¼=¼?-@-BžBžDEG€G€IñIñKbLbNÓNÓPDQDSµSµU&V&X—X—Z[]y]y_ê_êa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1û}!1Ðû}!1™¾!1L™¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1Íç!1(Íç!1k("1¤k("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1ŸQ"1€ŸQ"1=’"1ü=’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1CÚ"1´ CÚ"1£þ"10 £þ"1å+#1På+#1µ:#1,µ:#1‘‘#1˜‘‘#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1Õ€&1ÌÕ€&1¥R'1¥R'1Me'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘; b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 O�Þ‘Ñ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘; b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 O�Þ‘Ñ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘;  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 O�Þ‘Ñ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ¿ß{›& bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D„ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 �  � ��T” €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞTTÊÊ)@*@4¶4¶>,?,I I ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽzzððf f"Ü"Ü$R%R'È'È)>*>,´,´.*/*1 1 346Œ6Œ89;x;x=î=î?d@dBÚBÚDPEPGÆGÆI8?8I¬I¬S T ^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðxxððhhàà X X Ð ÐHHÀÀ88°°%%šš „ „"ù"ù$n%n'ã'ã)X*X,Í,Í.B/B1·1·3,4,6¡6¡89;‹;‹=>@u@uBêBêD_E_GÔGÔIHJHL¼L¼N0O0Q¤Q¤STVŒVŒXY[t[t]è]è_\`\bÐbÐdDeDg¸g¸i,j,l l noqˆqˆsüsüupvpxäxäzX{X}Ì}Ì@€@‚´‚´„(…(‡œ‡œ‰ŠŒ†Œ†ŽûŽû�p‘p“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1õ}!1�õ}!1“¾!1 “¾!1YÅ!1YÅ!1Çç!1€Çç!1e("1üe("1+/"1+/"1™Q"1p™Q"17’"1ì7’"1ý˜"1ôý˜"1Ù"1`Ù"1eý"1Ìeý"150#1ì50#1#1L #1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1#�&1T #�&1óR'1˜óR'1›e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒ+÷ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ÿß€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒ+÷ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ÿß€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒ+÷ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ÿß €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿ß  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss ê êaaØØ)O*O4Æ4Æ>;?;I°I°S%T%^š^šhis„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssêêaaØØ O O Æ Æ==´´++¢¢�� ~ ~"õ"õ$l%l'ã'ã)Z*Z,Ñ,Ñ.H/H1¿1¿36466­6­8$9$;›;›=>@‰@‰BCEwEwGîGîIcJcLØLØNMOMQÂQÂS7T7V¬V¬X!Y![–[–] ^ `€`€bõbõdjejgßgßiTjTlÉlÉn>o>q³q³s(t(v�v�xy{‡{‡}ü}üq€q‚æ‚æ„[…[‡ЇЉEŠEŒºŒºŽ/�/‘¤‘¤“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Å}!1�Å}!1c¾!1c¾!1¹Â!1$¹Â!1—ç!1�—ç!15("15("1‹,"1$‹,"1iQ"1�iQ"1’"1’"1]–"1$]–"1£Ú"1�£Ú"1ÿ"1üÿ"150#1 50#1#1| #1Õ$1Ü Õ$1­ò$1  ­ò$1çÞ%1¤ çÞ%1«€&1„ «€&1{R'1È{R'1#e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fe™þÊè´ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fe™þÊè´ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Fe™þÊè´ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1]}  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ì ìaaÖÖ)K*K4À4À>3?3I¦I¦ST^Š^Šhühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvììbbØØ N N Ä Ä::°°&&œœ††ûûp p"å"å$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ì=ì?a@aBÖBÖDKEKGÀGÀI3J3L¦L¦NOQŒQŒSÿSÿUrVrXåXåZX[X]Ë]Ë_>`>b°b°d"e"g”g”ijlxlxnênêp\q\sÎsÎu@v@x²x²z${$}–}–€‚z‚z„ì„ì†^‡^‰щыDŒDŽ·Ž·�*‘*“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1õ}!1�õ}!1“¾!1 “¾!1YÅ!1YÅ!1Çç!1€Çç!1e("1üe("1+/"1+/"1™Q"1p™Q"17’"1ì7’"1ý˜"1ôý˜"1Ù"1`Ù"1eý"1Ìeý"150#1ì50#1#1L #1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1½€&1T ½€&1�R'1˜�R'15e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒ-ù b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1žÞ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒ-ù b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1žÞ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oƒ-ù b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1žÞ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿ß  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ê ê``ÕÕ)J*J4¿4¿>2?2I¥I¥ST^‹^‹hþhþrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttêê``ÖÖ L L  Â88®®$$šš††üür r"è"è$]%]'Ò'Ò)G*G,¼,¼.1/11¦1¦346�6�89;z;z=ï=ï?d@dBÙBÙDNENGÃGÃI6J6L©L©NOQ�Q�STVuVuXèXèZ[[[]Î]Î_A`Ab´b´d'e'gšgši j l€l€nónópfqfsÙsÙuLvLx¿x¿z2{2}¥}¥€‚‹‚‹„þ„þ†q‡q‰ä‰ä‹WŒWŽÊŽÊ�=‘=“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Å}!1�Å}!1c¾!1 c¾!1³Â!1³Â!1—ç!1€—ç!15("1ü5("1…,"1…,"1iQ"1piQ"1’"1ì’"1W–"1ôW–"1—Ú"1`—Ú"1÷þ"1Ì÷þ"1/0#1ì/0#1û~#1L û~#1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜ÿË b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜ÿË b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜ÿË b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Qq  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ë ëbbÙÙ)P*P4Ç4Ç>>?>IµIµS,T,^ž^žhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttëëbbÙÙ P P Ç Ç>>µµ,,££‘‘  "ö"ö$m%m'ä'ä)[*[,Ò,Ò.I/I1À1À37476®6®8%9%;œ;œ=>@Š@ŠBCExExGïGïIfJfLÝLÝNTOTQËQËSBTBV¹V¹X0Y0[§[§]^`•`•b c eƒeƒgúgúiqjqlèlènZoZqÌqÌs>t>v°v°x"y"{”{”}~€x€x‚ê‚ê„\…\‡·Ή@Š@Œ²Œ²Ž$�$‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1û}!1�û}!1™¾!1 ™¾!1GÅ!1GÅ!1Çç!1€Çç!1e("1üe("1/"1/"1™Q"1p™Q"17’"1ì7’"1ë˜"1ôë˜"1ÁÚ"1`ÁÚ"1!ÿ"1Ì!ÿ"1/0#1ì/0#1û~#1L û~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1ù€&1T ù€&1ÉR'1˜ÉR'1qe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ4 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ѐ} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ4 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ѐ} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ4 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ð €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1|œ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss í íccÙÙ)O*O4Ã4Ã>7?7I«I«ST^“^“hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssííggáá [ [ Õ ÕOOÉÉCC½½33©© • •" # %�%�'÷'÷)m*m,ã,ã.Y/Y1Ï1Ï3E4E6»6»8191;§;§=>@�@�BCEwEwGëGëI_J_LÓLÓNGOGQ»Q»S/T/V£V£XY[‹[‹]ÿ]ÿ_s`sbçbçd[e[gÏgÏiCjCl·l·n+o+qŸqŸstv‡v‡xûxûzo{o}ã}ãW€W‚˂˄?…?‡±‡±‰$Š$Œ—Œ—Ž � ‘}‘}“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1¿}!1�¿}!1]¾!1 ]¾!1³Â!1³Â!1‘ç!1€‘ç!1/("1ü/("1…,"1…,"1cQ"1pcQ"1’"1ì’"1W–"1ôW–"1£Ú"1`£Ú"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1·€&1T ·€&1‡R'1˜‡R'1/e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–ûÇ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–ûÇ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–ûÇ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1^~  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ç ç\\ÑÑ)F*F4»4»>0?0I¥I¥ST^Œ^Œhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrçç\\ÑÑ F F » »00¥¥��yyîîc c"Ø"Ø$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=ß=ß?T@TBÉBÉD>E>G³G³I(J(L�L�NOQ‡Q‡SüSüUqVqXæXæZ[[[]Ð]Ð_E`Ebºbºd/e/g¤g¤ijlŽlŽnoqrqrsäsäuVvVxÈxÈz:{:}¬}¬€‚�‚�„…‡t‡t‰æ‰æ‹XŒXŽÊŽÊ�<‘<“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1û}!1�û}!1™¾!1 ™¾!1GÅ!1GÅ!1Íç!1€Íç!1k("1ük("1/"1/"1ŸQ"1pŸQ"1=’"1ì=’"1ë˜"1ôë˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1/0#1ì/0#1û~#1L û~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ6 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‘Ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ6 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‘Ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ6 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‘Ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1�¡  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss ë ë__ÓÓ)G*G4¹4¹>+?+I�I�ST^�^�hóhórcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssëëccÛÛ S S Ë ËCC»»33««““ { {"ï"ï$c%c'×'×)K*K,¿,¿.3/31§1§346�6�89;w;w=é=é?[@[BÍBÍD?E?G±G±I#J#L•L•NOQyQySëSëU]V]XÏXÏZA[A]³]³_%`%b—b—d e g{g{iíiík_l_nÑnÑpCqCsµsµu'v'x™x™z { }}}}ïï�a‚a„Ó„Ó†C‡C‰´‰´‹%Œ%Ž–Ž–�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1¿}!1�¿}!1]¾!1 ]¾!1¹Â!1¹Â!1‹ç!1€‹ç!1)("1ü)("1‹,"1‹,"1]Q"1p]Q"1û‘"1ìû‘"1]–"1ô]–"1‹Ú"1`‹Ú"1ëþ"1Ìëþ"1/0#1ì/0#1û~#1L û~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1�&1T �&1çR'1˜çR'1�e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•üÈ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•üÈ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•üÈ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Jj  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt î îhhÞÞ)T*T4Ê4Ê>>?>I²I²S&T&^š^šhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttîîhhââ \ \ Ö ÖPPÊÊDD¾¾88²²,, ¦ ¦" # %–%–' ( *‚*‚,ø,ø.n/n1ä1ä3Z4Z6Ð6Ð8F9F;¼;¼=2>2@¨@¨BCE”E”G H J~J~LòLòNfOfQÚQÚSNTNVÂVÂX6Y6[ª[ª]^`’`’bcezezgîgîibjblÖlÖnJoJq¾q¾s2t2v¦v¦xy{Ž{Ž}~€v€v‚ê‚ê„^…^‡Ò‡Ò‰FŠFŒºŒºŽ.�.‘¢‘¢“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1Ÿ¾!1 Ÿ¾!1;Å!1;Å!1Íç!1€Íç!1k("1ük("1 /"1 /"1ŸQ"1pŸQ"1=’"1ì=’"1ߘ"1ôߘ"1µÚ"1`µÚ"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1A�&1T A�&1S'1˜S'1¹e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•=  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ŒÌ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•=  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ŒÌ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•=  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ŒÌ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1u•  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo è èaaÚÚ)S*S4Ì4Ì>C?CIºIºS/T/^¢^¢his†s†} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooèèaaÚÚ S S Ì ÌEE¾¾77°°))¢¢ ” ”" # %†%†'ÿ'ÿ)x*x,ñ,ñ.j/j1ã1ã3\4\6Õ6Õ8N9N;Ç;Ç=@>@@¹@¹B2C2E«E«G$H$J›J›LMO‰O‰QRTwTwVîVîXeYe[Ü[Ü]S^S`Ê`Êb?c?e´e´g)h)jžjžlmo†o†qùqùsltlvßvßxRyR{Å{Å}8~8€«€«‚ƒ…‘…‘‡ˆŠtŠtŒæŒæŽX�X‘ʑʓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1¹}!1�¹}!1W¾!1 W¾!1¹Â!1¹Â!1‹ç!1€‹ç!1)("1ü)("1‹,"1‹,"1]Q"1p]Q"1û‘"1ìû‘"1]–"1ô]–"1�Ú"1`�Ú"1ýþ"1Ìýþ"1/0#1ì/0#1û~#1L û~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“øÄ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“øÄ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“øÄ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1]}  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ê êbbÖÖ)J*J4¾4¾>0?0I¢I¢ST^†^†høhørjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrêêbbÚÚ R R Ê ÊBBºº22ªª""šš Š Š"#%v%v'ê'ê)^*^,Ò,Ò.F/F1º1º3.4.6¢6¢89;Š;Š=þ=þ?r@rBæBæDZEZGÎGÎI@J@L²L²N$O$Q–Q–STVzVzXìXìZ^[^]Ð]Ð_B`Bb´b´d&e&g˜g˜i j l|l|nînîp`q`sÒsÒuDvDx¶x¶z({(}š}š € ‚~‚~„ð„ð†b‡b‰Ô‰Ô‹FŒFޏޏ�*‘*“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1Ÿ¾!1 Ÿ¾!1;Å!1;Å!1Óç!1€Óç!1q("1üq("1 /"1 /"1¥Q"1p¥Q"1C’"1ìC’"1ߘ"1ôߘ"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Ï€&1T Ï€&1ŸR'1˜ŸR'1Ge'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•?  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‹Ë€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•?  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‹Ë€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•?  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‹Ë €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Šª  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo æ æ]]ÔÔ)K*K4Â4Â>7?7I¬I¬ST^�^�hisqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooææ]]ÔÔ K K  Â99°°''žžŒŒ z z"ñ"ñ$h%h'ß'ß)V*V,Í,Í.D/D1»1»32426©6©8 9 ;—;—=>@…@…BüBüDsEsGêGêI_J_LÔLÔNIOIQ¾Q¾S3T3V¨V¨XY[’[’]^`|`|bïbïdbebgÕgÕiHjHl»l»n,o,q�q�stvvxðxðza{a}Ò}ÒC€C‚´‚´„%…%‡•‡•‰ŠŒwŒwŽèŽè�Y‘Y“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1¹}!1�¹}!1W¾!1 W¾!1³Â!1³Â!1…ç!1€…ç!1#("1ü#("1…,"1…,"1WQ"1pWQ"1õ‘"1ìõ‘"1W–"1ôW–"1ùØ"1`ùØ"1Yý"1ÌYý"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ùÅ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1F€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ùÅ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1F€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ùÅ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1F €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿ß  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ã ãXXÍÍ)B*B4µ4µ>(?(I›I›ST^�^�hôhôrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““}}òòg g"Ü"Ü$Q%Q'Æ'Æ);*;,°,°.%/%1š1š346„6„8ù8ù:n;n=á=á?T@TBÇBÇD:E:G­G­I J L“L“NOQyQySìSìU_V_XÒXÒZE[E]¸]¸_+`+bžbždeg„g„i÷i÷kjljnÝnÝpPqPsÃsÃu6v6x©x©z{}�}�€‚u‚u„è„è†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1~!1�~!1Ÿ¾!1 Ÿ¾!1GÅ!1GÅ!1Óç!1€Óç!1q("1üq("1/"1/"1¥Q"1p¥Q"1C’"1ìC’"1ë˜"1ôë˜"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1)0#1ì)0#1õ~#1L õ~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1M�&1T M�&1S'1˜S'1Åe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžF b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1˜Ø€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžF b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1˜Ø€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžF b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1˜Ø €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1p�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk ã ã[[ÓÓ)K*K4Ã4Ã>9?9I¯I¯S%T%^›^›his‡s‡} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkãã[[ÓÓ K K à Ã;;³³++££““  ƒ ƒ"û"û$s%s'ë'ë)c*c,Û,Û.S/S1Ë1Ë3C4C6»6»8393;«;«=#>#@›@›BCE‹E‹GHJyJyLïLïNeOeQÛQÛSQTQVÇVÇX=Y=[³[³])^)`Ÿ`Ÿbce‹e‹ghjwjwlílíncocqÙqÙsOtOvÅvÅx;y;{±{±}'~'€�€�‚ƒ…‰…‰‡ÿ‡ÿ‰uŠuŒëŒëŽa�a‘בד b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1³}!1�³}!1Q¾!1 Q¾!1³Â!1³Â!1…ç!1€…ç!1#("1ü#("1…,"1…,"1WQ"1pWQ"1õ‘"1ìõ‘"1W–"1ôW–"1ùØ"1`ùØ"1Yý"1ÌYý"1)0#1ì)0#1õ~#1L õ~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%13€&1T 3€&1R'1˜R'1«d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�õÁ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1G€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�õÁ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1G€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�õÁ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1G €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿ß  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ã ãVVÉÉ)<*<4­4­>?I�I�ST^q^qhâhârQsQ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““yyìì_ _"Ò"Ò$E%E'¸'¸)+*+,ž,ž./1„1„3÷3÷5j6j8Ý8Ý:P;P=Á=Á?2@2B£B£DEG…G…IöIöKgLgNØNØPIQISºSºU+V+XœXœZ [ ]~]~_ï_ïa`b`dÑdÑfBgBi³i³k$l$n•n•pqswswuèuèwYxYzÊzÊ|;};¬¬�‚„ބކý†ýˆm‰m‹Ý‹Ý�MŽM�½�½’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1~!1�~!1¥¾!1 ¥¾!1GÅ!1GÅ!1Óç!1€Óç!1q("1üq("1/"1/"1¥Q"1p¥Q"1C’"1ìC’"1ë˜"1ôë˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1)0#1ì)0#1õ~#1L õ~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Õ€&1T Õ€&1¥R'1˜¥R'1Me'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžH b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1™Ù€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžH b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1™Ù€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžH b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1™Ù €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1‰©  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk á áWWÍÍ)C*C4¹4¹>-?-I¡I¡ST^‰^‰hýhýrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkááWWÍÍ C C ¹ ¹//¥¥‘‘}}óói i"ß"ß$U%U'Ë'Ë)A*A,·,·.-/-1£1£346�6�89;{;{=ñ=ñ?g@gBÝBÝDSESGÉGÉI=J=L±L±N%O%Q™Q™S T V�V�XõXõZi[i]Ý]Ý_Q`QbÅbÅd9e9g­g­i!j!l•l•n o q}q}sñsñuevexÙxÙzM{M}Á}Á5€5‚©‚©„…‡‘‡‘‰ŠŒyŒyŽíŽí�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1­}!1�­}!1K¾!1 K¾!1¹Â!1¹Â!1ç!1€ç!1("1ü("1‹,"1‹,"1QQ"1pQQ"1ï‘"1ìï‘"1]–"1ô]–"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Õ€&1T Õ€&1¥R'1˜¥R'1Me'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ö b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿ß  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ä äYYÎÎ)C*C4¸4¸>,?,I I ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllää\\ÔÔ L L Ä Ä<<´´,,¤¤ŽŽ x x"í"í$b%b'×'×)L*L,Á,Á.6/61«1«3 4 6•6•8 9 ;;=ô=ô?i@iBÞBÞDSESGÈGÈI/?/I¤I¤ST^Ž^Žhisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbébéd^e^gÓgÓiHjHl½l½n2o2q§q§stv‘v‘xy{{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1­}!1�­}!1K¾!1 K¾!1¹Â!1¹Â!1yç!1€yç!1("1ü("1‹,"1‹,"1KQ"1pKQ"1é‘"1ìé‘"1]–"1ô]–"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ò¾ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ò¾ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ò¾ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿ß  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj à àUUÊÊ)?*?4´4´>'?'IšIšS T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjààVVÌÌ B B ¸ ¸..¤¤��zzïïd d"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¤b¤degˆgˆiúiúklllnÞnÞpPqPsÂsÂu4v4x¦x¦z{}Š}Šüü�n‚n„à„à†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1 ~!1� ~!1«¾!1 «¾!1/Å!1/Å!1Ùç!1€Ùç!1w("1üw("1/"1/"1«Q"1p«Q"1I’"1ìI’"1Ó˜"1ôÓ˜"1ÁÚ"1`ÁÚ"1!ÿ"1Ì!ÿ"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§Q b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‰É€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§Q b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‰É€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§Q b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‰É €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Šª  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞTTÉÉ)>*>4³4³>&?&I™I™S T ^^hòhòrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽzzððf f"Ü"Ü$Q%Q'Æ'Æ);*;,°,°.%/%1š1š346„6„8ù8ù:n;n=ã=ã?X@XBÍBÍDBEBG·G·I*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b¨b¨degŽgŽijltltnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1§}!1�§}!1E¾!1 E¾!1§Â!1§Â!1yç!1€yç!1("1ü("1y,"1y,"1KQ"1pKQ"1é‘"1ìé‘"1K–"1ôK–"1óØ"1`óØ"1Sý"1ÌSý"1#0#1ì#0#1ï~#1L ï~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ó€&1T ó€&1ÃR'1˜ÃR'1ke'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒó¿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1C€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒó¿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1C€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒó¿ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1C €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿ß  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>2?2I©I©S T ^’^’hisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÑLÑNHOHQ¿Q¿S6T6V­V­X$Y$[›[›]^`‰`‰bcewewgîgîiejelÜlÜnNoNqÀqÀs2t2v¤v¤xy{ˆ{ˆ}ú}úl€l‚Þ‚Þ„P…P‡‡‰4Š4Œ¦Œ¦Ž�‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1 ~!1� ~!1«¾!1 «¾!1;Å!1;Å!1ßç!1€ßç!1}("1ü}("1 /"1 /"1±Q"1p±Q"1O’"1ìO’"1ߘ"1ôߘ"1óØ"1`óØ"1Sý"1ÌSý"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°X$ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1”Ô€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°X$ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1”Ô€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°X$ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1”Ô €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿ß  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á áWWÍÍ)C*C4·4·>+?+IŸIŸST^‡^‡hûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±''�� ‰ ‰"ÿ"ÿ$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@ƒ@ƒB÷B÷DkEkGßGßISJSLÇLÇN;O;Q¯Q¯S#T#V—V—X Y [[]ó]ó_g`gbÛbÛdOeOgÃgÃi7j7l«l«noq“q“stv{v{xïxïzc{c}×}×K€K‚¿‚¿„3…3‡¥‡¥‰ŠŒ‹Œ‹ŽþŽþ�q‘q“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1§}!1�§}!1E¾!1 E¾!1§Â!1§Â!1sç!1€sç!1("1ü("1y,"1y,"1EQ"1pEQ"1ã‘"1ìã‘"1K–"1ôK–"1óØ"1`óØ"1Sý"1ÌSý"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠï» b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1D€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠï» b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1D€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠï» b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1D €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿ß  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Û ÛPPÅÅ):*:4¯4¯>$?$I™I™ST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÛÛPPÅÅ : : ¯ ¯$$™™ƒƒøømmââW W"Ì"Ì$A%A'¶'¶)+*+, , ./1Š1Š3ÿ3ÿ5t6t8é8é:^;^=Ó=Ó?H@HB½B½D2E2G§G§IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1 ~!1� ~!1«¾!1 «¾!1;Å!1;Å!1ßç!1€ßç!1}("1ü}("1 /"1 /"1±Q"1p±Q"1O’"1ìO’"1ߘ"1ôߘ"1óØ"1`óØ"1Sý"1ÌSý"1#0#1ì#0#1ï~#1L ï~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1i€&1T i€&19R'1˜9R'1ád'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°Z& b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1•Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°Z& b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1•Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°Z& b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1•Õ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿ß  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßSSÇÇ);*;4­4­>?I‘I‘ST^u^uhçhçrWsW} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ‡‡ûûo o"ã"ã$W%W'Ë'Ë)?*?,³,³.'/'1›1›346ƒ6ƒ8÷8÷:k;k=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b‹b‹dýdýfogoiáiákSlSnÅnÅp7q7s©s©uvx�x�zÿzÿ|q}qãã�U‚U„DŽdž7‡7‰¨‰¨‹ŒŽŠŽŠ�û�û’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1¡}!1�¡}!1?¾!1 ?¾!1­Â!1­Â!1sç!1€sç!1("1ü("1,"1,"1EQ"1pEQ"1ã‘"1ìã‘"1Q–"1ôQ–"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ð¼ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ð¼ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ð¼ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿ß  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh â â\\ÒÒ)H*H4¾4¾>2?2I¦I¦ST^�^�hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²,,¦¦  š š"#%Š%Š'(*v*v,ì,ì.b/b1Ø1Ø3N4N6Ä6Ä8:9:;°;°=&>&@œ@œBCEˆEˆGþGþIrJrLæLæNZOZQÎQÎSBTBV¶V¶X*Y*[ž[ž]^`†`†búbúdnengâgâiVjVlÊlÊn@o@q´q´s(t(vœvœxy{„{„}ø}øl€l‚à‚à„T…T‡ȇȉ<Š<Œ°Œ°Ž$�$‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1~!1�~!1±¾!1 ±¾!1)Å!1)Å!1åç!1€åç!1ƒ("1üƒ("1û."1û."1·Q"1p·Q"1U’"1ìU’"1͘"1ô͘"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1/�&1T /�&1ÿR'1˜ÿR'1§e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹a- b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ˆÈ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹a- b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ˆÈ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹a- b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ˆÈ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿ß  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ü ÜUUÎÎ)G*G4À4À>7?7I®I®S#T#^ ^ his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÜÜUUÎÎ G G À À99²²++¤¤–– ˆ ˆ"#%z%z'ó'ó)l*l,å,å.^/^1×1×3P4P6É6É8B9B;»;»=4>4@­@­B&C&EŸEŸGHJ�J�LMO}O}QôQôSkTkVâVâXYYY[Ð[Ð]G^G`¾`¾b3c3e¨e¨ghj’j’lmo„o„q÷q÷sjtjvÝvÝxPyP{Ã{Ã}6~6€©€©‚ƒ…�…�‡ˆŠrŠrŒäŒäŽV�V‘ȑȓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1›}!1�›}!19¾!1 9¾!1­Â!1­Â!1mç!1€mç!1 ("1ü ("1,"1,"1?Q"1p?Q"1Ý‘"1ìÝ‘"1Q–"1ôQ–"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ì¸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ì¸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ì¸ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿ß  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÊÊ)>*>4²4²>$?$I–I–ST^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$j%j'Þ'Þ)R*R,Æ,Æ.:/:1®1®3"4"6–6–8 9 ;~;~=ò=ò?f@fBÚBÚDNENGÂGÂI4J4L¦L¦NOQŠQŠSüSüUnVnXàXàZR[R]Ä]Ä_6`6b¨b¨degŒgŒiþiþkplpnænæpXqXsÊsÊuŒ>ްް�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1~!1�~!1±¾!1 ±¾!1)Å!1)Å!1åç!1€åç!1ƒ("1üƒ("1û."1û."1·Q"1p·Q"1U’"1ìU’"1͘"1ô͘"1íØ"1`íØ"1Mý"1ÌMý"10#1ì0#1é~#1L é~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1½€&1T ½€&1�R'1˜�R'15e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹c/ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ˆÈ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹c/ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ˆÈ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹c/ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ˆÈ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿ß  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>+?+I I ST^Œ^Œhýhýrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞISJSLÈLÈN=O=Q²Q²S'T'VœVœXY[†[†]û]û_p`pbãbãdVeVgÉgÉi?I‘I‘ST^ƒ^ƒhöhörgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=×=×?J@JB½B½D0E0G£G£IJL‰L‰NüNüPoQoSâSâUUVUXÈXÈZ;[;]®]®_!`!b b deg†g†iùiùklllnßnßpRqRsÅsÅu8v8x«x«z{}‘}‘€‚w‚w„ê„ê†[‡[‰͉͋?Œ?ޱޱ�#‘#“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1~!1�~!1·¾!1 ·¾!1/Å!1/Å!1åç!1€åç!1ƒ("1üƒ("1/"1/"1·Q"1p·Q"1U’"1ìU’"1Ó˜"1ôÓ˜"1Ú"1`Ú"1þ"1Ôþ"10#1ô0#1ã~#1T ã~#1Û$1´ Û$1ñ$1x ñ$1áÀ%1| áÀ%1G�&1\ G�&1=Q'1 =Q'1åc'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂj6 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‘Ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂj6 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‘Ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂj6 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‘Ñ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1«ó`  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ × ×OOÑÑ)I*I4Á4Á>7?7I­I­S#T#^¥^¥his‘s‘} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__××OOÇÇ ? ? · ·//§§——‡‡ÿÿw w"ï"ï$q%q'é'é)a*a,Ù,Ù.Q/Q1É1É3A4A6¹6¹8191;©;©=!>!@™@™BCE‰E‰GHJwJwLíLíNcOcQÙQÙSOTOVÅVÅX;Y;[±[±]'^'`�`�bce‰e‰gÿgÿiujulëlënmomqãqãsYtYvÏvÏxEyE{»{»}1~1€§€§‚ƒ…“…“‡ ˆ ŠŠŒõŒõŽk�k‘á‘á“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1•}!1�•}!13¾!1 3¾!1›Â!1›Â!1gç!1€gç!1("1ü("1m,"1m,"19Q"1p9Q"1ב"1ìב"1?–"1: ?–"1ÉØ"1* ÉØ"1)ý"1 )ý"10#1r 0#1ã~#1J ã~#1Ÿ$1" Ÿ$1Çð$1 Çð$1¥À%1â ¥À%1'€&1º '€&1P'1Z P'1Åb'1z €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„éµ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1A€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„éµ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1A€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P„éµ b-w* operation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z Š ¢ ² Ê Ú ò   reasonOperatorKernelvalue6kernelSpecúPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })zoutputElements*1152zinput0Elements241472zinput1Elements*9216binvocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1A €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1¿»  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×NNÁÁ)4*44¥4¥>?I‡I‡ST^u^uhæhærUsU} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡þþqqääW W"Ê"Ê$=%='°'°)#*#,–,–. / 1|1|3ï3ï5b6b8Õ8Õ:H;H=¹=¹?*@*B›B›D E G}G}IîIîK_L_NÐNÐPAQAS²S²U#V#X”X”Z[]v]v_ç_çadbddÕdÕfFgFi·i·k(l(n™n™p q s{s{uìuìw]x]zÎzÎ|?}?°°�!‚!„’„’†‡‰q‰q‹á‹á�QŽQ�Á�Á’ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1~!1�~!1·¾!1 ·¾!1/Å!1/Å!1ëç!1€ëç!1‰("1ü‰("1/"1/"1½Q"1p½Q"1[’"1ì[’"1Ó˜"1ôÓ˜"1gÚ"1`gÚ"1Çþ"1ÔÇþ"10#1ô0#1ã~#1T ã~#1Ã$1´ Ã$1ëð$1x ëð$1ÉÀ%1| ÉÀ%1Ï€&1\ Ï€&1ÅP'1 ÅP'1mc'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂl8 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1’Ò€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂl8 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1’Ò€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂl8 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1’Ò €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1ͧ[  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Õ ÕKKÉÉ)?*?4µ4µ>)?)I�I�ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……ûûqqçç] ]"Ó"Ó$Q%Q'Ç'Ç)=*=,³,³.)/)1Ÿ1Ÿ346‹6‹89;w;w=í=í?c@cBÙBÙDOEOGÅGÅI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noq…q…sùsùumvmxáxázU{U}É}É=€=‚±‚±„%…%‡™‡™‰ Š Œ�Œ�ŽõŽõ�i‘i“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„+i!1Y}!1Y}!1÷½!1|÷½!1[Ã!1„[Ã!1‘Æ!1ð‘Æ!1Ð!1|Ð!1QÓ!1HQÓ!1+ç!1Ô+ç!1É'"1PÉ'"1--"1X--"1c0"1Äc0"1í9"1Pí9"1#="1#="1ýP"1¨ýP"1›‘"1$ ›‘"1ÿ–"1, ÿ–"15š"1˜ 5š"1¿£"1$ ¿£"1õ¦"1ð õ¦"17Ô"1| 7Ô"1aÚ"1XaÚ"1Áþ"1ÔÁþ"1,#1ô,#1‘:#1Б:#1Mo#1\Mo#1-�#1€-�#1+š#1¬+š#1É$18É$1¡ò$1ü¡ò$1ÛÞ%1ÛÞ%1Uì%1àUì%1á€&1lá€&1±R'1°±R'1Ye'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¹… b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Á bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¹… b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Á  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¹…  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Á bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u•:Zú bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1T\œø8–Ö4t  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d”\ÜÝ]8¸¦&”oïðpÌL¨(„`à>¾œúz Ø X ¶ 6 ”  r ò ¤ ä P Ð ,lÈf¦D bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa § §QQÑÑ)G*G4½4½>1?1I¥I¥S%T%^™^™h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÛÛUUÏÏ I I � �  ƒƒýýww!!——  ƒ ƒ"ù"ù$y%y'ï'ï)e*e,Û,Û.Q/Q1Ç1Ç3=4=6³6³8)9);Ÿ;Ÿ=>@‰@‰BýBýDqEqGåGåIYJYLÍLÍNAOAQµQµS)T)V�V�XY[…[…]ù]ù_m`mbíbídaeagÕgÕiIjIl½l½n1o1q¥q¥stv�v�xy{u{u}é}é]€]‚тфE…E‡¹‡¹‰,Š,ŒŸŒŸŽ�‘…‘…“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!11~!1è1~!1Ͼ!1dϾ!1}Å!1l}Å!1Ï!1èÏ!1=Ò!1´=Ò!1ýç!1@ýç!1›("1¼›("1O/"1ÄO/"1Ù8"1@Ù8"1<"1 <"1ÏQ"1˜ÏQ"1m’"1m’"1!™"1 !™"1«¢"1˜ «¢"1á¥"1d á¥"1#Ó"1ð #Ó"1yÚ"1Ì yÚ"1Ùþ"1H Ùþ"1,#1h,#1Í:#1DÍ:#1‰o#1Љo#1Y�#1ôY�#1W›#1 W›#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1�€&1à�€&1QR'1$QR'1ùd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ªÞˆT b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘}½â" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ªÞˆT b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘}½â" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ªÞˆT  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘}½â"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 •µSs:Z bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Zl¬WÄq±  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dšlìéiW׸8B‡-­˜ƒînÙYÆF³3   Î  � z ú g ç T Ô AÁ,lׄÄ1q bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×NN��)8*84­4­> ? I“I“ST^…^…høhørksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡þþssèè] ]"Ò"Ò$%'ˆ'ˆ)ý)ý+r,r.ç.ç0�1�346z6z8ï8ï:d;d=Ù=Ù?L@LB¿B¿D2E2G¥G¥IJL‹L‹NþNþPqQqSäSäUWVWXÊXÊZ=[=]°]°_#`#b¢b¢degˆgˆiûiûknlnnánápTqTsÇsÇu:v:x­x­z { }“}“€‚y‚y„ì„ì†_‡_‰щыCŒC޵޵�'‘'“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„+i!1Y}!1Y}!1÷½!1|÷½!1[Ã!1„[Ã!1‘Æ!1ð‘Æ!1Ð!1|Ð!1QÓ!1HQÓ!1%ç!1Ô%ç!1Ã'"1PÃ'"1--"1X--"1c0"1Äc0"1í9"1Pí9"1#="1#="1÷P"1¨÷P"1•‘"1$ •‘"1ÿ–"1, ÿ–"15š"1˜ 5š"1¿£"1$ ¿£"1õ¦"1ð õ¦"17Ô"1| 7Ô"1+Ú"1X+Ú"1‹þ"1Ô‹þ"1Í+#1ôÍ+#1-;#1Ð-;#1éo#1\éo#19’#1€9’#17�#1¬7�#1±$18±$1‰ò$1ü‰ò$1ÃÞ%1ÃÞ%1=ì%1à=ì%1i€&1li€&19R'1°9R'1ád'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¶‚ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Á bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¶‚ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Á  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¶‚  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Á bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 f† )ü bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1…Å�ÝJŠ MÐ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d…Š �”8¸�™ÊJž“£#¨(«+ ® . ± 1 ´ 4 · 7 - m º :½=À@ÅŠÊM�P bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa ¤ ¤MMÉÉ)=*=4±4±>#?#I•I•ST^…^…h÷h÷risi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A „ „üüttììdd  ��õõi i"Ý"Ý$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=á=á?S@SBÅBÅD7E7G©G©IJL�L�NÿNÿPqQqSãSãUUVUXÇXÇZ9[9]«]«_`b›b›d e ggiñiñkclcnÕnÕpGqGs¹s¹u+v+x�x�z{}�}�óó�e‚e„ׄ׆I‡I‰º‰º‹+Œ+ŽœŽœ� ‘ “# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!11~!1è1~!1Ͼ!1dϾ!1}Å!1l}Å!1Ï!1èÏ!1=Ò!1´=Ò!1è!1@è!1¡("1¼¡("1O/"1ÄO/"1Ù8"1@Ù8"1<"1 <"1ÕQ"1˜ÕQ"1s’"1s’"1!™"1 !™"1«¢"1˜ «¢"1á¥"1d á¥"1#Ó"1ð #Ó"1CÚ"1Ì CÚ"1£þ"1H £þ"1å+#1hå+#13;#1D3;#1ïo#1Ðïo#1K’#1ôK’#1I�#1 I�#1Ÿ$1¬Ÿ$1wò$1pwò$1±Þ%1t±Þ%1+ì%1T+ì%1!€&1à!€&1ñQ'1$ñQ'1™d'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¬àŒX b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘~¾â" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¬àŒX b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘~¾â" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¬àŒX  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘~¾â"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u• @ - bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1†Æ©éLŒPÔ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d†Œ ’˜VÖ%¥žôt¤$ÃCª*°0´4 ¸ 8 ¼ < ] � À @ Ä D È HÌLÐPÖœÜ` $d bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×NNŒŒ)4*44§4§>?I‰I‰ST^w^whèhèrYsY} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡þþqqääW W"Ê"Ê$%'{'{)î)î+a,a.Ô.Ô0|1|3ï3ï5b6b8Õ8Õ:H;H=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éafbfd×d×fHgHi¹i¹k*l*n›n›p q s}s}uîuîw_x_zÐzÐ|A}A²²�#‚#„”„”†‡‰u‰u‹å‹å�UŽU�Å�Å’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1›}!1è›}!19¾!1d9¾!1ƒÅ!1lƒÅ!1 Ï!1è Ï!1CÒ!1´CÒ!1gç!1@gç!1("1¼("1U/"1ÄU/"1ß8"1@ß8"1<"1 <"19Q"1˜9Q"1ב"1ב"1'™"1 '™"1±¢"1˜ ±¢"1ç¥"1d ç¥"1)Ó"1ð )Ó"1Ú"1Ì Ú"1ßþ"1H ßþ"1!,#1h!,#1s:#1Ds:#1/o#1Ð/o#1Ž#1ôŽ#1}™#1 }™#1Ï$1¬Ï$1§ò$1p§ò$1áÞ%1táÞ%1[ì%1T[ì%1)�&1à)�&1ùR'1$ùR'1¡e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…îº b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  LQ‘â" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…îº b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  LQ‘â" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…îº  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  LQ‘â"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 v–üYy bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1WS“æ&{»P  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d—SÓâb&¦­-ùyxøÌLŸròEŘímÂB— l ì A Á  – ë k À @ Ç  “ Ó &f»ûP� bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ú ÚSSÌÌ)E*E4¾4¾>7?7IzIzS#T#^ ^ his†s†} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÚÚSSÌÌ E E ¾ ¾77°°))¢¢””  † †"ÿ"ÿ$x%x'ñ'ñ)j*j,ã,ã.\/\1Õ1Õ3N4N6Ç6Ç8@9@;¹;¹=2>2@«@«B$C$E�E�GHJ�J�LMO}O}QôQôSkTkV®V®X%Y%[œ[œ]^`Š`Šb3c3e¨e¨ghj’j’lmo„o„q÷q÷sjtjvÝvÝxPyP{Ã{Ã}6~6€©€©‚ƒ…�…�‡ˆŠtŠtŒæŒæŽX�X‘ʑʓ  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1%~!1è%~!1þ!1dþ!1ƒÅ!1lƒÅ!1 Ï!1è Ï!1CÒ!1´CÒ!1÷ç!1@÷ç!1•("1¼•("1U/"1ÄU/"1ß8"1@ß8"1<"1 <"1ÉQ"1˜ÉQ"1g’"1g’"1'™"1 '™"1±¢"1˜ ±¢"1ç¥"1d ç¥"1)Ó"1ð )Ó"1—Ú"1Ì —Ú"1÷þ"1H ÷þ"19,#1h9,#1¯:#1D¯:#1ko#1Ðko#1«�#1ô«�#1©š#1 ©š#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1À&1àÀ&1“R'1$“R'1;e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŸÓ}I b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘£ãâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŸÓ}I b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘£ãâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŸÓ}I  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘£ãâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ”´p�8X bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1 Mb¢D¨èLŒ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d �bâÚZDħ'&¦tôˆêjÌL®.�tôXØ< ¼     „ è h Ì L Ý  ° 0’Ò4tØ|¼ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜQQÆÆ);*;4°4°>&?&IšIšST^P^Phøhørnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœ††ûûp p"å"å$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ì=ì?a@aBÖBÖDKEKGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`JbÀbÀd4e4g¨g¨ijl�l�nÐnÐpDqDs¸s¸u,v,x x zH{H}¼}¼0€0‚¤‚¤„…‡އމŠŒxŒxŽíŽí�b‘b“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1•}!1è•}!13¾!1d3¾!1ƒÅ!1lƒÅ!1 Ï!1è Ï!1CÒ!1´CÒ!1gç!1@gç!1("1¼("1U/"1ÄU/"1ß8"1@ß8"1<"1 <"19Q"1˜9Q"1ב"1ב"1'™"1 '™"1±¢"1˜ ±¢"1ç¥"1d ç¥"1)Ó"1ð )Ó"1UÚ"1Ì UÚ"1µþ"1H µþ"1÷+#1h÷+#1g:#1Dg:#1#o#1Ð#o#1CŽ#1ôCŽ#1A™#1 A™#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1·€&1à·€&1‡R'1$‡R'1/e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ì¸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  KQ‘â" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ì¸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  KQ‘â" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…ì¸  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  KQ‘â"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 k‹,L& bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Q‘¢ââ"s³D  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dQÑ"¢ósÄDCÕŽfæÙY7·¤$ˆÙYª*{ûL Ì  � î n ¿ ? �  a ¡ R ’ ò 2ƒÃT bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>+?+IkIkST^Œ^Œhýhýroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜISJSLÈLÈN=O=Q²Q²S'T'VgVgXÜXÜZQ[Q]Æ]Æ_;`;bãbãdVeVgÉgÉi!?!I”I”S T ^G^Ghîhîrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?M@MBÂBÂD7E7G¬G¬I!J!L”L”NOQzQzSíSíU`V`XÓXÓZF[F]¹]¹_,`,b¢b¢deg†g†iøiøkjljn§n§pqs‹s‹uýuýwoxoz{}ˆ}ˆúú�l‚l„Þ„Þ†R‡R‰ʼnŋ8Œ8ޫޫ�‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1›}!1Л}!19¾!1L9¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1mç!1(mç!1 ("1¤ ("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1?Q"1€?Q"1Ý‘"1üÝ‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#1o2#1,o2#1 Œ#1˜ Œ#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%15�&1Ì5�&1S'1S'1­e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ï» b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TQ‘â" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ï» b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TQ‘â" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†ï»  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TQ‘â"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w—þt” bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1P€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�Ý]ª*w÷à  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ù ÙQQÉÉ)A*A4¹4¹>1?1I§I§ST^“^“h i ss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A ¹ ¹11©©!!™™‰‰ y y"ñ"ñ$i%i'á'á)Y*Y,Ñ,Ñ.I/I1Á1Á39496±6±8)9);¡;¡=>@‘@‘B C E�E�GùGùIqJqLçLçN]O]QÓQÓSITIV¿V¿X5Y5[«[«]!^!`—`—b c eƒeƒgùgùiojolålån[o[qÑqÑsGtGv½v½x3y3{©{©}~€•€•‚ ƒ …�…�‡÷‡÷‰mŠmŒãŒãŽY�Y‘ϑϓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1~!1Ð~!1½¾!1L½¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1ñç!1(ñç!1�("1¤�("1[/"1¬[/"1å8"1(å8"1<"1ô<"1ÃQ"1€ÃQ"1a’"1üa’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1»Ú"1´ »Ú"1ÿ"10 ÿ"1],#1P],#1]2#1,]2#1Ù‹#1˜Ù‹#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1í€&1Ìí€&1½R'1½R'1ee'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”Èr> b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘‡Çæ& bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”Èr> b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘‡Çæ& bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”Èr>  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘‡Çæ&  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 “³6V�¯ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ËK–aá° ð  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>0?0I§I§ST^�^�histst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIXJXLÏLÏNFOFQ½Q½S4T4V«V«X"Y"[™[™]^`‡`‡bþbþdueugìgìicjclÚlÚnLoLq¾q¾s0t0v¢v¢xy{†{†}ø}øj€j‚܂܄N…N‡À‡À‰2Š2Œ¤Œ¤Ž�‘ˆ‘ˆ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1›}!1Л}!19¾!1L9¾!1ƒÅ!1TƒÅ!1 Ï!1Ð Ï!1CÒ!1œCÒ!1mç!1(mç!1 ("1¤ ("1U/"1¬U/"1ß8"1(ß8"1<"1ô<"1?Q"1€?Q"1Ý‘"1üÝ‘"1'™"1 '™"1±¢"1€ ±¢"1ç¥"1L ç¥"1)Ó"1Ø )Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1ý4#1,ý4#1™Œ#1˜™Œ#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†í¹ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UQ‘â" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†í¹ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UQ‘â" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†í¹  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UQ‘â"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p�Qq0 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Ý€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1DÄ‘Þ^«+m ­  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×MMÃÃ)9*94¯4¯>%?%I™I™S T ^�^�hõhõrisi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$K%K'Á'Á)7*7,­,­.#/#1™1™346…6…8û8û:q;q=ç=ç?]@]BÓBÓDIEIG¿G¿I5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`Ib½b½d1e1g¥g¥ijl�l�noququséséu]v]xÑxÑzE{E}¹}¹-€-‚¡‚¡„…‡‰‡‰‰ý‰ý‹qŒqŽåŽå�Y‘Y“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l1i!1%~!1Ð%~!1þ!1Lþ!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1÷ç!1(÷ç!1•("1¤•("1[/"1¬[/"1å8"1(å8"1<"1ô<"1ÉQ"1€ÉQ"1g’"1üg’"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1ë4#1,ë4#1iŒ#1˜iŒ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–ÊvB b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘ˆÈç' bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–ÊvB b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘ˆÈç' bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–ÊvB  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘ˆÈç'  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 s“1hˆ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1‹Ë€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1,¬÷wÂB� ; {  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb × ×LLÁÁ)6*64«4«>"?"I—I—S T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbb××LLÁÁ 6 6 « «  ••  ôôiiÞÞS S"È"È$=%='²'²)'*',œ,œ./1†1†3û3û5p6p8å8å:Z;Z=Ï=Ï?D@DB¹B¹D.E.G£G£IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1¡}!1С}!1?¾!1L?¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1sç!1(sç!1("1¤("1[/"1¬[/"1å8"1(å8"1<"1ô<"1EQ"1€EQ"1ã‘"1üã‘"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1�Ú"1´ �Ú"1ýþ"10 ýþ"1?,#1P?,#1i2#1,i2#1'Œ#1˜'Œ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1ÿ€&1Ìÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ð¼ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 AS“ä$ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ð¼ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 AS“ä$ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ð¼  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 AS“ä$  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x˜ o� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1G€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1‡ÖV¥%tô÷ 7  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>+?+I I ST^Š^Šhÿhÿrtst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜISJSLÈLÈN=O=Q²Q²S'T'VœVœXY[†[†]û]û_p`pbåbådZeZgÏgÏiDjDl¹l¹n.o.q£q£stv�v�xy{w{w}ì}ìa€a‚Ö‚Ö„K…K‡À‡À‰5Š5ŒªŒªŽ�‘”‘”“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1~!1Ð~!1·¾!1L·¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ëç!1(ëç!1‰("1¤‰("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1½Q"1€½Q"1[’"1ü[’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1»Ú"1´ »Ú"1ÿ"10 ÿ"1],#1P],#1c2#1,c2#1�Œ#1˜�Œ#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ÿ€&1Ìÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½g3 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘˜Øñ1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½g3 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘˜Øñ1 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½g3  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘˜Øñ1  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ’²4TŠª bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÖV¬,‚P �  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh â â\\ÒÒ)H*H4¾4¾>4?4I¨I¨ST^�^�hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²,,¦¦  š š"#%Š%Š'(*v*v,ì,ì.b/b1Ø1Ø3N4N6Ä6Ä8:9:;°;°=&>&@œ@œBCEˆEˆGþGþItJtLèLèN\O\QÐQÐSDTDV¸V¸X,Y,[ [ ]^`ˆ`ˆbübüdpepgägäiXjXlÌlÌn@o@q´q´s(t(vœvœxy{„{„}ø}øl€l‚à‚à„T…T‡ȇȉ<Š<Œ°Œ°Ž$�$‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1¡}!1С}!1?¾!1L?¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1sç!1(sç!1("1¤("1[/"1¬[/"1å8"1(å8"1<"1ô<"1EQ"1€EQ"1ã‘"1üã‘"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1÷4#1,÷4#1·Œ#1˜·Œ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡îº b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @R’ã# bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡îº b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @R’ã# bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡îº  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @R’ã#  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u•:Xx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1–Ö€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1CÃ’áa°0† Æ  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÃÃ)8*84­4­>"?"I•I•ST^{^{hîhîrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD*>4²4²>&?&I˜I˜S T ^|^|hîhîr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$j%j'Þ'Þ)R*R,Æ,Æ.:/:1®1®3"4"6–6–8 9 ;~;~=ò=ò?f@fBÚBÚDNENGÂGÂI6J6L¨L¨NOQŒQŒSþSþUpVpXâXâZT[T]Æ]Æ_8`8bªbªdegŽgŽijlrlrnänäpVqVsÈsÈu:v:x¬x¬z{}�}�€‚t‚t„æ„æ†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1§}!1Ч}!1E¾!1LE¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1yç!1(yç!1("1¤("1/"1¬/"1 9"1( 9"1?<"1ô?<"1KQ"1€KQ"1é‘"1üé‘"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"11Ú"1´ 1Ú"1‘þ"10 ‘þ"1Ó+#1PÓ+#1i2#1,i2#1uŒ#1˜uŒ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1á€&1Ìá€&1±R'1±R'1Ye'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆñ½ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Ii©@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆñ½ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Ii©@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆñ½  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Ii©@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ááy™# bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÓS¦&yù@ €  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee ß ßUUËË)A*A4·4·>+?+IŸIŸST^‡^‡hûhûroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯%%›› ‡ ‡"ý"ý$s%s'é'é)_*_,Õ,Õ.K/K1Á1Á37476­6­8#9#;™;™=>@ƒ@ƒB÷B÷DkEkGßGßISJSLÇLÇN;O;Q¯Q¯S#T#V—V—X Y [[]ó]ó_g`gbÛbÛdOeOgÃgÃi7j7l«l«noq“q“stv{v{xïxïzc{c}×}×K€K‚¿‚¿„3…3‡§‡§‰ŠŒ�Œ�Ž�‘s‘s“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1~!1Ð~!1±¾!1L±¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1åç!1(åç!1ƒ("1¤ƒ("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1·Q"1€·Q"1U’"1üU’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1aÚ"1´ aÚ"1Áþ"10 Áþ"1,#1P,#1i2#1,i2#15�#1˜5�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²\( b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘…Åå% bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²\( b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘…Åå% bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²\(  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘…Åå%  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‘±Áá2R bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€áaÂB£#@ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛPPÅÅ):*:4¯4¯>"?"I•I•ST^{^{hîhîrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹uuêê_ _"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?N@NBÁBÁD4E4G§G§IJL�L�NOQsQsSæSæUYVYXÌXÌZ?[?]²]²_%`%b˜b˜d e g~g~iñiñkdldn×n×pJqJs½s½u0v0x£x£z{}‰}‰üü�o‚o„â„â†U‡U‰ljNj9Œ9ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1§}!1Ч}!1E¾!1LE¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1sç!1(sç!1("1¤("1y/"1¬y/"19"1(9"19<"1ô9<"1EQ"1€EQ"1ã‘"1üã‘"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1ý4#1,ý4#1�#1˜�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1i€&1Ìi€&19R'19R'1ád'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆï» b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jh¨þ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆï» b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jh¨þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆï»  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jh¨þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ããzš$D bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1“Ó€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1LÌŸòrÅEÓ  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝQQÅÅ)9*94­4­>?I‘I‘ST^u^uhçhçrYsY} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��……ùùm m"á"á$U%U'É'É)=*=,±,±.%/%1™1™3 4 6�6�8õ8õ:i;i=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b‹b‹dýdýfogoiáiákSlSnÅnÅp7q7s©s©uvx�x�zÿzÿ|q}qãã�U‚U„DŽdž9‡9‰ª‰ª‹ŒŽŒŽŒ�ý�ý’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1~!1Ð~!1·¾!1L·¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1åç!1(åç!1ƒ("1¤ƒ("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1·Q"1€·Q"1U’"1üU’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#1 5#1, 5#1×�#1˜×�#1¥$1$¥$1}ò$1è}ò$1·Þ%1ì·Þ%1'€&1Ì'€&1÷Q'1÷Q'1Ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´`, b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘…Åæ& bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´`, b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘…Åæ& bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´`,  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘…Åæ&  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 t”Áá8 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¡á€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1„eåFÆ'§¡á €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛNNÁÁ)4*44§4§>?I‰I‰SúSú]k^khÜhÜrMsM} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹þþqqääW W"Ê"Ê$=%='°'°)#*#,–,–. / 1|1|3ï3ï5b6b8Õ8Õ:H;H=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éaZbZdËdËf;?;I²I²S'T'^š^šh i s€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÞÞWWÐÐ I I  Â;;´´--¦¦˜˜ Š Š"#%|%|'õ'õ)n*n,ç,ç.`/`1Ù1Ù3R4R6Ë6Ë8D9D;½;½=6>6@¯@¯B(C(E¡E¡GHJ“J“L M O�O�QøQøSoToVæVæX]Y][Ô[Ô]K^K`Â`Âb7c7e¬e¬g!h!j–j–l m o~o~qñqñsdtdv×v×xJyJ{½{½}0~0€£€£‚ƒ…‰…‰‡ü‡ü‰nŠnŒàŒàŽR�R‘đē b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1 ~!1Ð ~!1«¾!1L«¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1Ùç!1(Ùç!1w("1¤w("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1«Q"1€«Q"1I’"1üI’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1[Ú"1´ [Ú"1»þ"10 »þ"1ý+#1Pý+#1i2#1,i2#1é�#1˜é�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1Õ€&1ÌÕ€&1¥R'1¥R'1Me'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§Q b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘‘Ñã# bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§Q b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘‘Ñã# bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§Q  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘‘Ñã#  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Áጬ(H bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ìlØXÄD°ð €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àUUÊÊ)?*?4´4´>*?*IžIžST^†^†húhúrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  ŠŠÿÿt t"é"é$^%^'Ó'Ó)H*H,½,½.2/21§1§346‘6‘89;{;{=ð=ð?e@eBÚBÚDOEOGÄGÄI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡�‡�‰ŠŒzŒzŽïŽï�d‘d“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1­}!1Э}!1K¾!1LK¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1yç!1(yç!1("1¤("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1KQ"1€KQ"1é‘"1üé‘"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1aÚ"1´ aÚ"1Áþ"10 Áþ"1,#1P,#1 5#1, 5#1Å�#1˜Å�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ð¼ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @R’ã# bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ð¼ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @R’ã# bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ð¼  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @R’ã#  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÁáŸ.N bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1žÞ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1xøVÖ4´’ŽÎ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>/?/I¤I¤ST^ˆ^ˆhùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbçbçdZeZgÍgÍi@j@l³l³n$o$q•q•stvwvwxèxèzY{Y}Ê}Ê;€;‚¬‚¬„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1 ~!1Ð ~!1«¾!1L«¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1ßç!1(ßç!1}("1¤}("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1±Q"1€±Q"1O’"1üO’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#15#1,5#1‘Ž#1˜‘Ž#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©U! b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘‘Ñã# bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©U! b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘‘Ñã# bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©U!  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘‘Ñã#  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ááw—> bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¬ì€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1°0œˆtô\œ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜQQÆÆ);*;4°4°>%?%I˜I˜S T ^|^|hîhîrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒvvëë` `"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1³}!1г}!1Q¾!1LQ¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ç!1(ç!1("1¤("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1QQ"1€QQ"1ï‘"1üï‘"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1CÚ"1´ CÚ"1£þ"10 £þ"1å+#1På+#1o2#1,o2#1Ý�#1˜Ý�#1á$1$á$1¹ò$1è¹ò$1óÞ%1ìóÞ%1G�&1ÌG�&1S'1S'1¿e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠó¿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @R’ã# bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠó¿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @R’ã# bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠó¿  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @R’ã#  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Áá{› - bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ëkÔT½=¢â €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝUUÍÍ)E*E4½4½>5?5I«I«S!T!^—^—h i sƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%���� } }"õ"õ$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=>@•@•B C E…E…GýGýIuJuLëLëNaOaQ×Q×SMTMVÃVÃX9Y9[¯[¯]%^%`›`›bce‡e‡gýgýisjslélén_o_qÕqÕsKtKvÁvÁx7y7{­{­}#~#€™€™‚ƒ…………‡û‡û‰qŠqŒçŒçŽ]�]‘Ó‘Ó“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1~!1Ð~!1¥¾!1L¥¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1Óç!1(Óç!1q("1¤q("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1¥Q"1€¥Q"1C’"1üC’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1CÚ"1´ CÚ"1£þ"10 £þ"1å+#1På+#1o2#1,o2#1�Ž#1˜�Ž#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1ÿ€&1Ìÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœF b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘‡Çâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœF b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘‡Çâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœF  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘‡Çâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÁáŸ. bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€÷wînåe`  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>4?4I«I«S"T"^”^”hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãI\J\LÓLÓNJOJQÁQÁS8T8V¯V¯X&Y&[�[�]^`‹`‹bceyeygðgðigjglÞlÞnPoPqÂqÂs4t4v¦v¦xy{Š{Š}ü}ün€n‚à‚à„R…R‡ćĉ6Š6Œ¨Œ¨Ž�‘Œ‘Œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1­}!1Э}!1K¾!1LK¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ç!1(ç!1("1¤("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1QQ"1€QQ"1ï‘"1üï‘"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1mÚ"1´ mÚ"1Íþ"10 Íþ"1,#1P,#15#1,5#1…Ž#1˜…Ž#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠñ½ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @R’ã# bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠñ½ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @R’ã# bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠñ½  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @R’ã#  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Áá‚¢4T bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1©é€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¦&�xøaáK‹ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÇÇ)=*=4³4³>)?)I�I�ST^…^…hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒuŒuŽéŽé�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1~!1Ð~!1¥¾!1L¥¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1Ùç!1(Ùç!1w("1¤w("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1«Q"1€«Q"1I’"1üI’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#1'5#1,'5#1Q�#1˜Q�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1¥€&1Ì¥€&1uR'1uR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžJ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘†Æâ" bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžJ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘†Æâ" bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžJ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘†Æâ"  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Áár’4 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1·÷€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ü\ÓSÊJÁAW €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Û ÛPPÅÅ):*:4¯4¯>&?&I›I›ST^‚^‚hôhôrfsf} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÛÛPPÅÅ : : ¯ ¯$$™™ƒƒøømmââW W"Ì"Ì$A%A'¶'¶)+*+, , ./1Š1Š3ÿ3ÿ5t6t8é8é:^;^=Ó=Ó?H@HB½B½D2E2G§G§IJL“L“NOQ}Q}SòSòUgVgXÜXÜZQ[Q]Æ]Æ_;`;b°b°d%e%gšgšijl„l„nönöphqhsÚsÚuLvLx¾x¾z0{0}¢}¢€‚†‚†„ø„ø†j‡j‰܉܋NŒNŽÀŽÀ�2‘2“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1¹}!1й}!1W¾!1LW¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1…ç!1(…ç!1#("1¤#("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1WQ"1€WQ"1õ‘"1üõ‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"17Ú"1´ 7Ú"1—þ"10 —þ"1Ù+#1PÙ+#1u2#1,u2#1‘Ž#1˜‘Ž#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1�&1Ì�&1áR'1áR'1‰e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ôÀ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @R’ã# bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ôÀ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @R’ã# bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ôÀ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @R’ã#  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Áár’ý bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ôtèhÜ\P� €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>/?/I¤I¤ST^Ž^Žhisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbébéd^e^gÓgÓiHjHl½l½n2o2q§q§stv‘v‘xy{{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1û}!1Ðû}!1™¾!1L™¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1Íç!1(Íç!1k("1¤k("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1ŸQ"1€ŸQ"1=’"1ü=’"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"11Ú"1´ 1Ú"1‘þ"10 ‘þ"1Ó+#1PÓ+#1u2#1,u2#1Q�#1˜Q�#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1�&1Ì�&1çR'1çR'1�e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘; b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘‘Ñã# bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘; b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘‘Ñã# bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘;  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘‘Ñã#  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Áár’ô bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€‚„†P €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll æ æ``ÖÖ)L*L4Â4Â>8?8I¬I¬S T ^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllææ``ÚÚ T T Î ÎHHÂÂ<<¶¶00ªª$$ ž ž"#%Ž%Ž'(*z*z,ð,ð.f/f1Ü1Ü3R4R6È6È8>9>;´;´=*>*@ @ BCEŒEŒGHJxJxLìLìN`O`QÔQÔSHTHV¼V¼X0Y0[¤[¤]^`Œ`Œbcetetgègèi\j\lÐlÐnDoDq¸q¸s,t,v v xy{ˆ{ˆ}ü}üp€p‚ä‚ä„X…X‡̇̉@Š@Œ´Œ´Ž(�(‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1³}!1г}!1Q¾!1LQ¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1…ç!1(…ç!1#("1¤#("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1WQ"1€WQ"1õ‘"1üõ‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1aÚ"1´ aÚ"1Áþ"10 Áþ"1,#1P,#1'5#1,'5#1E�#1˜E�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1«€&1Ì«€&1{R'1{R'1#e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ò¾ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @R’ã# bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ò¾ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @R’ã# bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ò¾  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @R’ã#  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Áá{›&F bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1´ô€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÐPÄD¸8¬,D €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜRRÇÇ)<*<4±4±>&?&I™I™S T ^^hòhòrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒxxîîd d"Ú"Ú$O%O'Ä'Ä)9*9,®,®.#/#1˜1˜3 4 6‚6‚8÷8÷:l;l=á=á?V@VBËBËD@E@GµGµI*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b¨b¨degŽgŽijltltnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1~!1Ð~!1Ÿ¾!1LŸ¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1Óç!1(Óç!1q("1¤q("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1¥Q"1€¥Q"1C’"1üC’"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1=Ú"1´ =Ú"1�þ"10 �þ"1ß+#1Pß+#135#1,35#1�#1˜�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1Ÿ€&1ÌŸ€&1oR'1oR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“?  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Q‘‘Ñã# bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“?  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Q‘‘Ñã# bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“?   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Q‘‘Ñã#  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Áái‰" bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ˆ Š ŒŽÒ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âZZÎÎ)B*B4¶4¶>*?*IœIœST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢’’  ‚ ‚"ú"ú$n%n'â'â)V*V,Ê,Ê.>/>1²1²3&4&6š6š89;‚;‚=ö=ö?j@jBÞBÞDRERGÆGÆI:J:L¬L¬NOQ�Q�STVtVtXæXæZX[X]Ê]Ê_<`v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Ñ}!1�Ñ}!1o¾!1o¾!1¹Â!1$¹Â!1£ç!1�£ç!1A("1A("1‹,"1$‹,"1uQ"1�uQ"1’"1’"1]–"1$]–"1£Ú"1�£Ú"1ÿ"1üÿ"150#1 50#1#1| #1ç$1Ü ç$1¿ò$1  ¿ò$1ùÞ%1¤ ùÞ%1�&1„ �&1ÕR'1ÈÕR'1}e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ai�Òß« b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ai�Òß« b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ai�Òß« b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1]}  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10zz ô ôjjàà)V*V4Ê4Ê>>?>I²I²S&T&^š^šhis€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðzzôônnèè b b Ü ÜVVÐÐJJÄÄ::°°&& œ œ"#%ˆ%ˆ'þ'þ)t*t,ê,ê.`/`1Ö1Ö3L4L6Â6Â8898;®;®=">"@–@–B C E~E~GòGòIfJfLÚLÚNNONQÂQÂS6T6VªVªXY[’[’]^`z`zbîbîdbebgÖgÖiJjJl¾l¾n2o2q¦q¦stvŽvŽxy{v{v}ê}ê^€^‚Ò‚Ò„F…F‡¸‡¸‰+Š+ŒžŒžŽ�‘„‘„“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1û}!1�û}!1™¾!1 ™¾!1YÅ!1YÅ!1Çç!1€Çç!1e("1üe("1+/"1+/"1™Q"1p™Q"17’"1ì7’"1ý˜"1ôý˜"1ÍÚ"1`ÍÚ"1-ÿ"1Ì-ÿ"150#1ì50#1#1L #1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…/û b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ý€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…/û b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ý€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…/û b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ý €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1‚¢  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr è è]]ÒÒ)G*G4º4º>-?-I I ST^†^†hùhùrjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrèè^^ÔÔ J J À À66¬¬""˜˜  ‚‚÷÷l l"á"á$V%V'Ë'Ë)@*@,µ,µ.*/*1Ÿ1Ÿ346‰6‰8þ8þ:s;s=æ=æ?Y@YBÌBÌD?E?G²G²I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b£b£deg‰g‰iüiükolonânâpUqUsÈsÈu;v;x®x®z!{!}”}”€‚z‚z„í„í†^‡^‰ЉЋBŒBŽ´Ž´�&‘&“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Ë}!1�Ë}!1i¾!1i¾!1¹Â!1$¹Â!1�ç!1��ç!1;("1;("1‹,"1$‹,"1oQ"1�oQ"1 ’"1 ’"1]–"1$]–"1¯Ú"1�¯Ú"1ÿ"1üÿ"150#1 50#1#1| #1Ï$1Ü Ï$1§ò$1  §ò$1áÞ%1¤ áÞ%1�€&1„ �€&1]R'1È]R'1e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bg›Îâ® b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bg›Îâ® b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Bg›Îâ® b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1d„  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx ð ðddØØ)L*L4¾4¾>0?0I¢I¢ST^†^†høhørhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðxxððhhàà X X Ð ÐHHÀÀ88°°$$˜˜  € €"ô"ô$h%h'Ü'Ü)P*P,Ä,Ä.8/81¬1¬3 4 6”6”89;|;|=î=î?`@`BÒBÒDDEDG¶G¶I(J(LšLšN O Q~Q~SðSðUbVbXÔXÔZF[F]¸]¸_*`*bœbœdeg€g€iòiòkdldnÖnÖpHqHsºsºu,v,xžxžz{}‚}‚ôô�f‚f„؄؆H‡H‰¹‰¹‹*Œ*ޛޛ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1û}!1�û}!1™¾!1 ™¾!1SÅ!1SÅ!1Íç!1€Íç!1k("1ük("1%/"1%/"1ŸQ"1pŸQ"1=’"1ì=’"1÷˜"1ô÷˜"1ÍÚ"1`ÍÚ"1-ÿ"1Ì-ÿ"150#1ì50#1#1L #1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1E€&1T E€&1R'1˜R'1½d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…1ý b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1œÜ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…1ý b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1œÜ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Q…1ý b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1œÜ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ƒ£  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ê ê]]ÐÐ)C*C4´4´>%?%I–I–ST^x^xhéhérXsX} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttêê``ÖÖ L L  Â88®®$$šš  €€óóf f"Ù"Ù$L%L'¿'¿)2*2,¥,¥./1‹1‹3þ3þ5q6q8ä8ä:W;W=È=È?9@9BªBªDEGŒGŒIýIýKnLnNßNßPPQPSÁSÁU2V2X£X£Z[]…]…_ö_öagbgdØdØfIgIiºiºk+l+nœnœp q s~s~uïuïw`x`zÑzÑ|B}B³³�$‚$„•„•†‡‰t‰t‹ä‹ä�TŽT�Ä�Ä’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ë}!1�Ë}!1i¾!1 i¾!1³Â!1³Â!1—ç!1€—ç!15("1ü5("1…,"1…,"1iQ"1piQ"1’"1ì’"1W–"1ôW–"1�Ú"1`�Ú"1ýþ"1Ìýþ"1/0#1ì/0#1û~#1L û~#1ó$1¬ ó$1Ëò$1p Ëò$1ß%1t ß%1M�&1T M�&1S'1˜S'1Åe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÏ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÏ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÏ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Vv  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ï ïhháá)Z*Z4Ó4Ó>J?JIÁIÁS6T6^©^©his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvïïhháá Z Z Ó ÓLLÅÅ>>··00©©"" › ›"#%�%�'(**,ø,ø.q/q1ê1ê3c4c6Ü6Ü8U9U;Î;Î=G>G@À@ÀB9C9E²E²G+H+J¢J¢LMO�O�QRT~T~VõVõXlYl[ã[ã]Z^Z`Ñ`ÑbFcFe»e»g0h0j¥j¥lmo�o�qrtstsvævæxYyY{Ì{Ì}?~?€²€²‚%ƒ%…˜…˜‡ ˆ Š{Š{ŒíŒíŽ_�_‘ёѓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1Ÿ¾!1 Ÿ¾!1GÅ!1GÅ!1Íç!1€Íç!1k("1ük("1/"1/"1ŸQ"1pŸQ"1=’"1ì=’"1ë˜"1ôë˜"1ÁÚ"1`ÁÚ"1!ÿ"1Ì!ÿ"1/0#1ì/0#1û~#1L û~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1ç€&1T ç€&1·R'1˜·R'1_e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ8 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‘Ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ8 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‘Ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ8 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‘Ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1}�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq é é^^ÓÓ)H*H4½4½>1?1I¥I¥ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqééaaÙÙ Q Q É ÉAA¹¹11©©““ } }"ò"ò$g%g'Ü'Ü)Q*Q,Æ,Æ.;/;1°1°3%4%6š6š89;„;„=ù=ù?n@nBãBãDXEXGÍGÍIAJALµLµN)O)Q�Q�STV…V…XùXùZm[m]á]á_U`UbÉbÉd=e=g±g±i%j%l™l™n o q�q�sõsõuivixÝxÝzQ{Q}Å}Å9€9‚­‚­„!…!‡•‡•‰ Š ŒŒŽôŽô�i‘i“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Å}!1�Å}!1c¾!1 c¾!1³Â!1³Â!1—ç!1€—ç!15("1ü5("1…,"1…,"1iQ"1piQ"1’"1ì’"1W–"1ôW–"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1/0#1ì/0#1û~#1L û~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜ÿË b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜ÿË b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜ÿË b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1e…  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ë ëbbÙÙ)P*P4Ç4Ç>>µµ,,££‘‘  "ö"ö$m%m'ä'ä)[*[,Ò,Ò.I/I1À1À37476®6®8%9%;œ;œ=>@Š@ŠBCExExGïGïIdJdLÙLÙNNONQÃQÃS8T8V­V­X"Y"[—[—] ^ `�`�bôbôdgeggÚgÚiMjMlÀlÀn1o1q¢q¢stv„v„xõxõzf{f}×}×H€H‚¹‚¹„*…*‡š‡š‰ Š Œ|Œ|ŽíŽí�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1Ÿ¾!1 Ÿ¾!1GÅ!1GÅ!1Óç!1€Óç!1q("1üq("1/"1/"1¥Q"1p¥Q"1C’"1ìC’"1ë˜"1ôë˜"1ÍÚ"1`ÍÚ"1-ÿ"1Ì-ÿ"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ: b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‘Ñ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ: b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‘Ñ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ: b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‘Ñ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Šª  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ç ç\\ÑÑ)F*F4»4»>.?.I¡I¡ST^…^…h÷h÷risi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqçç]]ÓÓ I I ¿ ¿55««!!——  ��öök k"à"à$U%U'Ê'Ê)?*?,´,´.)/)1ž1ž346ˆ6ˆ8ý8ý:r;r=ç=ç?\@\BÑBÑDFEFG»G»I.J.L¡L¡NOQ‡Q‡SúSúUmVmXàXàZS[S]Æ]Æ_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰̉̋?Œ?޲޲�%‘%“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Å}!1�Å}!1c¾!1 c¾!1¹Â!1¹Â!1‘ç!1€‘ç!1/("1ü/("1‹,"1‹,"1cQ"1pcQ"1’"1ì’"1]–"1ô]–"1‘Ú"1`‘Ú"1ñþ"1Ìñþ"1/0#1ì/0#1û~#1L û~#1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1Y�&1T Y�&1)S'1˜)S'1Ñe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ì b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ì b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ì b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Qq  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ê êbbÚÚ)R*R4Ê4Ê>@?@I¶I¶S,T,^¢^¢hisŽsŽ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrêêbbÚÚ R R Ê ÊBBºº22ªª""šš Š Š"#%z%z'ò'ò)j*j,â,â.Z/Z1Ò1Ò3J4J6Â6Â8:9:;²;²=*>*@¢@¢BCE’E’G H J€J€LöLöNlOlQâQâSXTXVÎVÎXDYD[º[º]0^0`¦`¦bce’e’ghj~j~lôlônjojqàqàsVtVvÌvÌxByB{¸{¸}.~.€¤€¤‚ƒ…�…�‡ˆŠ|Š|ŒòŒòŽh�h‘Þ‘Þ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1Ÿ¾!1 Ÿ¾!15Å!15Å!1Óç!1€Óç!1q("1üq("1/"1/"1¥Q"1p¥Q"1C’"1ìC’"1Ù˜"1ôÙ˜"1µÚ"1`µÚ"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—A  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1†Æ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—A  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1†Æ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—A  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1†Æ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1v–  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ä ä[[ÒÒ)I*I4À4À>7?7I®I®S%T%^—^—h i s{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmää[[ÒÒ I I À À77®®%%œœŠŠ x x"ï"ï$f%f'Ý'Ý)T*T,Ë,Ë.B/B1¹1¹30406§6§89;•;•= > @ƒ@ƒBúBúDqEqGèGèI_J_LÖLÖNMOMQÄQÄS;T;V²V²X)Y)[ [ ]^`Ž`Žbce|e|gógóijjjlálánSoSqÅqÅs7t7v©v©xy{�{�}ÿ}ÿq€q‚ã‚ã„U…U‡LJlj9Š9Œ«Œ«Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1¿}!1�¿}!1]¾!1 ]¾!1¿Â!1¿Â!1‘ç!1€‘ç!1/("1ü/("1‘,"1‘,"1cQ"1pcQ"1’"1ì’"1c–"1ôc–"1£Ú"1`£Ú"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1á€&1T á€&1±R'1˜±R'1Ye'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•üÈ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•üÈ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•üÈ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1d„  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æ\\ÒÒ)H*H4¾4¾>2?2I¦I¦ST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  ‚‚øøn n"ä"ä$Z%Z'Ð'Ð)F*F,¼,¼.2/21¨1¨346”6”8 9 ;€;€=ö=ö?l@lBâBâDXEXGÎGÎIBJBL¶L¶N*O*QžQžSTV†V†XúXúZn[n]â]â_V`VbÊbÊd>e>g²g²i&j&lšlšnoq‚q‚sösöujvjxÞxÞzR{R}Æ}Æ:€:‚®‚®„"…"‡–‡–‰ Š Œ~Œ~ŽòŽò�f‘f“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1¥¾!1 ¥¾!15Å!15Å!1Óç!1€Óç!1q("1üq("1/"1/"1¥Q"1p¥Q"1C’"1ìC’"1Ù˜"1ôÙ˜"1ÍÚ"1`ÍÚ"1-ÿ"1Ì-ÿ"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1±€&1T ±€&1�R'1˜�R'1)e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—C b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1…Å€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—C b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1…Å€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—C b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1…Å €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1‹«  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm â âWWÌÌ)A*A4¶4¶>+?+I I ST^‡^‡hùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmââWWÌÌ A A ¶ ¶++  ŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?O@OBÄBÄD9E9G®G®I#J#L˜L˜N O Q‚Q‚S÷S÷UlVlXáXáZV[V]Ë]Ë_@`@bµbµd*e*gŸgŸijl‰l‰nûnûpmqmsßsßuQvQxÃxÃz5{5}§}§€‚‹‚‹„ý„ý†o‡o‰á‰á‹SŒSŽÅŽÅ�7‘7“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1¹}!1�¹}!1W¾!1 W¾!1­Â!1­Â!1‹ç!1€‹ç!1)("1ü)("1,"1,"1]Q"1p]Q"1û‘"1ìû‘"1Q–"1ôQ–"1…Ú"1`…Ú"1åþ"1Ìåþ"1)0#1ì)0#1õ~#1L õ~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”ýÉ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1B€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”ýÉ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1B€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”ýÉ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1B €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Jj  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn å å\\ÓÓ)J*J4Á4Á>6?6I«I«S T ^•^•h i ss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnåå\\ÓÓ J J Á Á88¯¯&&��‹‹ y y"ð"ð$g%g'Þ'Þ)U*U,Ì,Ì.C/C1º1º31416¨6¨89;–;–= > @„@„BûBûDrErGéGéI^J^LÓLÓNHOHQ½Q½S2T2V§V§XY[‘[‘]^`{`{bðbðdeeegÚgÚiOjOlÄlÄn9o9q®q®s#t#v˜v˜x y {‚{‚}÷}÷l€l‚á‚á„V…V‡ˇˉ@Š@ŒµŒµŽ*�*‘Ÿ‘Ÿ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1~!1�~!1¥¾!1 ¥¾!1GÅ!1GÅ!1Ùç!1€Ùç!1w("1üw("1/"1/"1«Q"1p«Q"1I’"1ìI’"1ë˜"1ôë˜"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l J b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1šÚ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l J b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1šÚ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l J b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1šÚ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1q‘  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ç çaa××)M*M4Ã4Ã>7?7I«I«ST^“^“his{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmççaaÛÛ U U Ï ÏIIÃÃ==··11««%% Ÿ Ÿ"#%�%�'(*{*{,ñ,ñ.g/g1Ý1Ý3S4S6É6É8?9?;µ;µ=+>+@¡@¡BCE�E�GHJwJwLëLëN_O_QÓQÓSGTGV»V»X/Y/[£[£]^`‹`‹bÿbÿdsesgçgçi[j[lÏlÏnCoCq·q·s+t+vŸvŸxy{‡{‡}û}ûo€o‚ã‚ã„W…W‡ˇˉ?Š?Œ³Œ³Ž'�'‘›‘›“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1¹}!1�¹}!1W¾!1 W¾!1³Â!1³Â!1…ç!1€…ç!1#("1ü#("1…,"1…,"1WQ"1pWQ"1õ‘"1ìõ‘"1W–"1ôW–"1£Ú"1`£Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ùÅ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1C€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ùÅ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1C€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ùÅ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1C €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1e…  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ã ãYYÎÎ)C*C4¸4¸>+?+IžIžST^„^„h÷h÷rjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““  õõk k"á"á$V%V'Ë'Ë)@*@,µ,µ.*/*1Ÿ1Ÿ346‰6‰8þ8þ:s;s=è=è?]@]BÒBÒDGEGG¼G¼I/J/L¢L¢NOQˆQˆSûSûUnVnXáXáZT[T]Ç]Ç_:`:b­b­d e g“g“ijlylynìnìp_q_sÒsÒuEvEx¸x¸z+{+}ž}ž€‚„‚„„÷„÷†j‡j‰݉݋PŒPŽÃŽÃ�6‘6“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1 ~!1� ~!1«¾!1 «¾!1GÅ!1GÅ!1Ùç!1€Ùç!1w("1üw("1/"1/"1«Q"1p«Q"1I’"1ìI’"1ë˜"1ôë˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l L b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1›Û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l L b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1›Û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l L b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1›Û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Šª  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm å å]]ÑÑ)E*E4¹4¹>+?+I�I�ST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmåå]]ÕÕ M M Å Å==µµ--¥¥••  … …"ý"ý$q%q'å'å)Y*Y,Í,Í.A/A1µ1µ3)4)6�6�89;…;…=ù=ù?m@mBáBáDUEUGÉGÉI;J;L­L­NOQ‘Q‘STVuVuXçXçZY[Y]Ë]Ë_=`=b¯b¯d!e!g“g“ijlwlwnénép[q[sÍsÍu?v?x±x±z#{#}•}•€‚y‚y„ë„ë†]‡]‰ωϋAŒA޳޳�%‘%“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1³}!1�³}!1Q¾!1 Q¾!1¹Â!1¹Â!1…ç!1€…ç!1#("1ü#("1‹,"1‹,"1WQ"1pWQ"1õ‘"1ìõ‘"1]–"1ô]–"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1ç€&1T ç€&1·R'1˜·R'1_e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘úÆ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘úÆ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘úÆ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Áá  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn è è^^ÔÔ)J*J4¾4¾>2?2I¦I¦ST^Ž^Žhistst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnèèbbÜÜ V V Ð ÐJJÄÄ>>¸¸..¤¤ � �"#%|%|'ò'ò)h*h,Þ,Þ.T/T1Ê1Ê3@4@6¶6¶8,9,;¢;¢=>@Š@ŠBþBþDrErGæGæIZJZLÎLÎNBOBQ¶Q¶S*T*VžVžXY[†[†]ú]ú_n`nbâbâdVeVgÊgÊi>j>l²l²n&o&qšqšstv‚v‚xöxözj{j}Þ}ÞR€R‚ƂƄ:…:‡¬‡¬‰ŠŒ’Œ’Ž�‘x‘x“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1 ~!1� ~!1«¾!1 «¾!1/Å!1/Å!1ßç!1€ßç!1}("1ü}("1/"1/"1±Q"1p±Q"1O’"1ìO’"1Ó˜"1ôÓ˜"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©S b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1†Æ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©S b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1†Æ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©S b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1†Æ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Áá  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜQQÆÆ);*;4®4®>!?!I”I”ST^z^zhíhír^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒvvëë` `"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ú=Ú?M@MBÀBÀD3E3G¦G¦IJLŒLŒNÿNÿPrQrSåSåUXVXXËXËZ>[>]±]±_$`$b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†R‡R‰ĉċ6Œ6ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1³}!1�³}!1Q¾!1 Q¾!1¹Â!1¹Â!1ç!1€ç!1("1ü("1‹,"1‹,"1QQ"1pQQ"1ï‘"1ìï‘"1]–"1ô]–"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1o€&1T o€&1?R'1˜?R'1çd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�ö b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Áá  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ä äXXÌÌ)@*@4²4²>$?$I–I–ST^z^zhìhìr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllää\\ÔÔ L L Ä Ä<<´´,,¤¤ŒŒ t t"è"è$\%\'Ð'Ð)D*D,¸,¸.,/,1 1 346ˆ6ˆ8ü8ü:p;p=â=â?T@TBÆBÆD8E8GªGªIJLŽLŽNOQrQrSäSäUVVVXÈXÈZ:[:]¬]¬_`b�b�degtgtiæiækXlXnÊnÊp?IŠIŠSûSû]l^lhÝhÝrLsL} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽttççZ Z"Í"Í$@%@'³'³)&*&,™,™. / 113ò3ò5e6e8Ø8Ø:K;K=¼=¼?-@-BžBžDEG€G€IñIñKbLbNÓNÓPDQDSµSµU&V&X—X—Z[]y]y_ê_êa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1­}!1�­}!1K¾!1 K¾!1§Â!1§Â!1ç!1€ç!1("1ü("1y,"1y,"1QQ"1pQQ"1ï‘"1ìï‘"1K–"1ôK–"1óØ"1`óØ"1Sý"1ÌSý"1#0#1ì#0#1ï~#1L ï~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1/�&1T /�&1ÿR'1˜ÿR'1§e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ÷à b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ÷à b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ÷à b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Áá  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ã ã\\ÕÕ)N*N4Ç4Ç>>?>IµIµS*T*^�^�his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjãã\\ÕÕ N N Ç Ç@@¹¹22««$$�� � �"#%�%�'ú'ú)s*s,ì,ì.e/e1Þ1Þ3W4W6Ð6Ð8I9I;Â;Â=;>;@´@´B-C-E¦E¦GHJ–J–L M O„O„QûQûSrTrVéVéX`Y`[×[×]N^N`Å`Åb:c:e¯e¯g$h$j™j™lmo�o�qôqôsgtgvÚvÚxMyM{À{À}3~3€¦€¦‚ƒ…Œ…Œ‡ý‡ý‰oŠoŒáŒáŽS�S‘őœ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1~!1�~!1±¾!1 ±¾!1;Å!1;Å!1åç!1€åç!1ƒ("1üƒ("1 /"1 /"1·Q"1p·Q"1U’"1ìU’"1ߘ"1ôߘ"1óØ"1`óØ"1Sý"1ÌSý"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²\( b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1–Ö€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²\( b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1–Ö€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²\( b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1–Ö €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Áá  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝRRÇÇ)<*<4±4±>%?%I™I™S T ^�^�hõhõrisi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��‡‡üüq q"æ"æ$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`Ib½b½d1e1g¥g¥ijl�l�noququséséu]v]xÑxÑzE{E}¹}¹-€-‚¡‚¡„…‡‰‡‰‰þ‰þ‹sŒsŽèŽè�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1§}!1�§}!1E¾!1 E¾!1§Â!1§Â!1yç!1€yç!1("1ü("1y,"1y,"1KQ"1pKQ"1é‘"1ìé‘"1K–"1ôK–"1óØ"1`óØ"1Sý"1ÌSý"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1½€&1T ½€&1�R'1˜�R'15e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒó¿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒó¿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒó¿ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Áá  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>0?0I¥I¥ST^‰^‰húhúrjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n%o%q–q–stvxvxxéxézZ{Z}Ë}Ë<€<‚­‚­„…‡އމÿ‰ÿ‹pŒpŽáŽá�R‘R“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1~!1�~!1±¾!1 ±¾!1;Å!1;Å!1åç!1€åç!1ƒ("1üƒ("1 /"1 /"1·Q"1p·Q"1U’"1ìU’"1ߘ"1ôߘ"1óØ"1`óØ"1Sý"1ÌSý"1#0#1ì#0#1ï~#1L ï~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²^* b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1—×€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²^* b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1—×€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1~²^* b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1—× €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Áá  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛPPÅÅ):*:4¯4¯>"?"I•I•ST^y^yhëhër]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹uuêê_ _"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I"J"L•L•NOQ{Q{SîSîUaVaXÔXÔZG[G]º]º_-`-bŸbŸdegƒgƒiõiõkglgnÙnÙpKqKs½s½u/v/x¡x¡z{}…}…÷÷�i‚i„Û„Û†M‡M‰À‰À‹3Œ3ަަ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1§}!1�§}!1E¾!1 E¾!1­Â!1­Â!1yç!1€yç!1("1ü("1,"1,"1KQ"1pKQ"1é‘"1ìé‘"1Q–"1ôQ–"1óØ"1`óØ"1Sý"1ÌSý"10#1ì0#1é~#1L é~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1;�&1T ;�&1 S'1˜ S'1³e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ôÀ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ôÀ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ôÀ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Áá  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÎÎ)F*F4¾4¾>4?4IªIªS T ^˜^˜his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$n%n'æ'æ)^*^,Ö,Ö.N/N1Æ1Æ3>4>6¶6¶8.9.;¦;¦=>@–@–BCE†E†GþGþItJtLêLêN`O`QÖQÖSLTLVÂVÂX8Y8[®[®]$^$`š`šbce†e†gügüirjrlèlèn`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1~!1�~!1·¾!1 ·¾!1)Å!1)Å!1åç!1€åç!1ƒ("1üƒ("1û."1û."1·Q"1p·Q"1U’"1ìU’"1͘"1ô͘"1óØ"1`óØ"1Sý"1ÌSý"10#1ì0#1é~#1L é~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ó€&1T ó€&1ÃR'1˜ÃR'1ke'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»e1 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1…Å€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»e1 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1…Å€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»e1 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1…Å €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Áá  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>+?+I¢I¢ST^•^•hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜISJSLÊLÊNAOAQ¸Q¸S/T/V¦V¦XY[”[”] ^ `‚`‚bùbùdpepgçgçi^j^lÕlÕnQoQqÃqÃs5t5v§v§xy{‹{‹}ý}ýo€o‚á‚á„S…S‡Ňʼn7Š7Œ©Œ©Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1¡}!1�¡}!1?¾!1 ?¾!1­Â!1­Â!1sç!1€sç!1("1ü("1,"1,"1EQ"1pEQ"1ã‘"1ìã‘"1Q–"1ôQ–"1óØ"1`óØ"1Sý"1ÌSý"10#1ì0#1é~#1L é~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ð¼ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ð¼ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ð¼ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Áá  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÆÆ)<*<4²4²>&?&IšIšST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$N%N'Ä'Ä):*:,°,°.&/&1œ1œ346ˆ6ˆ8þ8þ:t;t=ê=ê?`@`BÖBÖDLELGÂGÂI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1~!1�~!1·¾!1 ·¾!1)Å!1)Å!1ëç!1€ëç!1‰("1ü‰("1û."1û."1½Q"1p½Q"1[’"1ì[’"1͘"1ô͘"1óØ"1`óØ"1Sý"1ÌSý"10#1ì0#1é~#1L é~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»g3 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1…Å€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»g3 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1…Å€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»g3 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1…Å €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Áá  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ö ÖKKÀÀ)5*54ª4ª>?I”I”S T ^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÖÖKKÀÀ 5 5 ª ª””  ~~óóhhÝÝR R"Ç"Ç$<%<'±'±)&*&,›,›./1…1…3ú3ú5o6o8ä8ä:Y;Y=Î=Î?C@CB¸B¸D-E-G¢G¢IJLŒLŒNOQvQvSëSëU`V`XÕXÕZJ[J]¿]¿_4`4b©b©deg“g“ijl}l}n÷n÷piqisÛsÛuMvMx¿x¿z1{1}£}£€‚‡‚‡„ù„ù†k‡k‰݉݋OŒOŽÁŽÁ�3‘3“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L1i!1¡}!1�¡}!1?¾!1 ?¾!1›Â!1›Â!1mç!1€mç!1 ("1ü ("1m,"1m,"1?Q"1p?Q"1Ý‘"1ìÝ‘"1?–"1ô?–"11Ú"1`1Ú"1‘þ"1Ô‘þ"10#1ô0#1ã~#1T ã~#1Ï$1´ Ï$1÷ð$1x ÷ð$1ÕÀ%1| ÕÀ%1�&1\ �&1ûP'1 ûP'1£c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆñ½ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆñ½ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆñ½ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1Öf·#  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÉÉ)@*@4·4·>,?,I¡I¡ST^—^—h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbæbæd[e[gÐgÐiEjElºlºn;o;q°q°s%t%všvšxy{„{„}ù}ùn€n‚ã‚ã„X…X‡͇͉BŠBŒ·Œ·Ž,�,‘¡‘¡“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1~!1�~!1½¾!1 ½¾!1/Å!1/Å!1ëç!1€ëç!1‰("1ü‰("1/"1/"1½Q"1p½Q"1[’"1ì[’"1Ó˜"1ôÓ˜"11Ú"1`1Ú"1‘þ"1Ô‘þ"10#1ô0#1ã~#1T ã~#1Õ$1´ Õ$1ýð$1x ýð$1ÛÀ%1| ÛÀ%1 �&1\  �&1Q'1 Q'1©c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Än: b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1”Ô€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Än: b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1”Ô€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Än: b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1”Ô €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1³ûo#  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Û ÛUUÕÕ)K*K4Á4Á>5?5I©I©ST^�^�his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÛÛUUÏÏ I I à Ã==··11««%%ŸŸ “ “" # %�%�'(*y*y,ï,ï.e/e1Û1Û3Q4Q6Ç6Ç8=9=;³;³=)>)@Ÿ@ŸBCE‹E‹GHJuJuLéLéN]O]QÑQÑSETEV¹V¹X-Y-[¡[¡]^`‰`‰býbýdqeqgågåiYjYlÍlÍnMoMqÁqÁs5t5v©v©xy{‘{‘}~€y€y‚í‚í„a…a‡Õ‡Õ‰IŠIŒ½Œ½Ž1�1‘¥‘¥“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1›}!1�›}!19¾!1 9¾!1›Â!1›Â!1mç!1€mç!1 ("1ü ("1m,"1m,"1?Q"1p?Q"1Ý‘"1ìÝ‘"1?–"1ô?–"1aÚ"1`aÚ"1Áþ"1ÔÁþ"10#1ô0#1ã~#1T ã~#1½$1´ ½$1åð$1x åð$1ÃÀ%1| ÃÀ%1Ÿ€&1\ Ÿ€&1•P'1 •P'1=c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†í¹ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†í¹ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1R†í¹ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1•æR  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×MMÆÆ);*;4°4°>#?#I–I–S T ^ˆ^ˆhûhûrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnðnðpcqcsÖsÖuIvIx¼x¼z/{/}¢}¢€‚ˆ‚ˆ„û„û†n‡n‰á‰á‹TŒTŽÇŽÇ�:‘:“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L1i!1~!1�~!1½¾!1 ½¾!15Å!15Å!1ñç!1€ñç!1�("1ü�("1/"1/"1ÃQ"1pÃQ"1a’"1ìa’"1Ù˜"1ôÙ˜"1yÚ"1`yÚ"1Ùþ"1ÔÙþ"10#1ô0#1ã~#1T ã~#1½$1´ ½$1åð$1x åð$1ÃÀ%1| ÃÀ%1“€&1\ “€&1‰P'1 ‰P'11c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Äp< b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1•Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Äp< b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1•Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Äp< b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1•Õ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1Ú"»o  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ù ÙQQÍÍ)A*A4µ4µ>'?'I™I™S T ^‰^‰hûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A ¹ ¹11©©!!™™‰‰ y y"ñ"ñ$m%m'á'á)U*U,É,É.=/=1±1±3%4%6™6™8 9 ;�;�=õ=õ?i@iBÝBÝDQEQGÅGÅI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnñnñpcqcsÕsÕuGvGx¹x¹z+{+}�}�€‚�‚�„ó„ó†e‡e‰׉׋IŒI޻޻�-‘-“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„+i!1_}!1_}!1ý½!1|ý½!1aÃ!1„aÃ!1—Æ!1ð—Æ!1!Ð!1|!Ð!1WÓ!1HWÓ!1+ç!1Ô+ç!1É'"1PÉ'"13-"1X3-"1i0"1Äi0"1ó9"1Pó9"1)="1)="1ýP"1¨ýP"1›‘"1$ ›‘"1—"1, —"1;š"1˜ ;š"1Å£"1$ Å£"1û¦"1ð û¦"1=Ô"1| =Ô"1[Ú"1X[Ú"1»þ"1Ô»þ"1ý+#1ôý+#1»:#1л:#1wo#1\wo#1 �#1€ �#1 ›#1¬ ›#1·$18·$1�ò$1ü�ò$1ÉÞ%1ÉÞ%1Cì%1àCì%1‡€&1l‡€&1WR'1°WR'1ÿd'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P»‡ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1à bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P»‡ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1à  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P»‡  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1à bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 „¤3S&F bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1g§N$d·÷`   bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dgçNÎ5µœƒêjÑQZÚ¸8#£¡!ìlŠ µ5 s ó \ Ü E Å . ®  — €´ ô ç'ŽÎ7wà  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10]] � �RRÇÇ)<*<4±4±>$?$I—I—ST^‰^‰hühüroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð]]ÓÓII¿¿ 5 5 u uëëaa××MMwwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?P@PBÃBÃD6E6G©G©IJL�L�NOQuQuSèSèU[V[XÎXÎZA[A]´]´_'`'b¦b¦degŒgŒiÿiÿkrlrnånåpXqXsËsËu>v>x±x±z${$}—}— € ‚}‚}„ð„ð†c‡c‰Õ‰Õ‹GŒG޹޹�+‘+“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!11~!1è1~!1Ͼ!1dϾ!1ƒÅ!1lƒÅ!1 Ï!1è Ï!1CÒ!1´CÒ!1è!1@è!1¡("1¼¡("1U/"1ÄU/"1ß8"1@ß8"1<"1 <"1ÕQ"1˜ÕQ"1s’"1s’"1'™"1 '™"1±¢"1˜ ±¢"1ç¥"1d ç¥"1)Ó"1ð )Ó"1Ú"1Ì Ú"1ßþ"1H ßþ"1!,#1h!,#1£:#1D£:#1_o#1Ð_o#1{�#1ô{�#1yš#1 yš#1Ã$1¬Ã$1›ò$1p›ò$1ÕÞ%1tÕÞ%1Oì%1TOì%1á€&1àá€&1±R'1$±R'1Ye'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¬àŒX b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“y¹æ& bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¬àŒX b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“y¹æ& bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¬àŒX  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“y¹æ&  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x˜ +\| bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1` @f¦¢âD„  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d`à@À  €à`À@  €  bâÿDÄÞ^ & ¦ ½ =  ˆ ê j Ì L ® . � p°PV–²òT” bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee ß ßUU——)C*C4¹4¹>-?-I¡I¡S!T!^•^•h i s}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯%%›› ‡ ‡"ý"ý$?%?'µ'µ)+*+,¡,¡./1Ã1Ã39496¯6¯8%9%;›;›=>@…@…BùBùDmEmGáGáIUJULÉLÉN=O=Q±Q±S%T%V™V™X Y [�[�]õ]õ_i`ibébéd]e]gÑgÑiEjEl¹l¹n-o-q¡q¡stv‰v‰xýxýzq{q}å}åY€Y‚͂̈́A…A‡µ‡µ‰(Š(Œ›Œ›Ž�‘�‘�“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„+i!1Y}!1Y}!1÷½!1|÷½!1aÃ!1„aÃ!1—Æ!1ð—Æ!1!Ð!1|!Ð!1WÓ!1HWÓ!1+ç!1Ô+ç!1É'"1PÉ'"13-"1X3-"1i0"1Äi0"1ó9"1Pó9"1)="1)="1ýP"1¨ýP"1›‘"1$ ›‘"1—"1, —"1;š"1˜ ;š"1Å£"1$ Å£"1û¦"1ð û¦"1=Ô"1| =Ô"1%Ú"1X%Ú"1…þ"1Ô…þ"1Ç+#1ôÇ+#1W;#1ÐW;#1p#1\p#1“#1€“#1ž#1¬ž#1Ÿ$18Ÿ$1wò$1üwò$1±Þ%1±Þ%1+ì%1à+ì%1'€&1l'€&1÷Q'1°÷Q'1Ÿd'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¸„ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1à bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¸„ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1à  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¸„  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1à bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 r’"ô bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Ð` .n­í ü<  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d�  °0À@ÐPà`ðp€ Ž ~ þ  œ G Ç * ª  � 8 ¸ Ù Y F Æ TÔ=}bâpð€ÀP�^ì, bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ ž žPPÃÃ)6*64©4©>?I‹I‹ST^y^yhêhêr[s[} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 v vììbbØØNNssææY Y"Ì"Ì$?%?'²'²)%*%,˜,˜. / 1~1~3ñ3ñ5d6d8×8×:J;J=½=½?.@.BŸBŸDEG�G�IòIòKcLcNÔNÔPEQES¶S¶U'V'X˜X˜Z [ ]z]z_ë_ëahbhdÙdÙfJgJi»i»k,l,n�n�pqssuðuðwaxazÒzÒ|C}C´´�%‚%„–„–†‡‰w‰w‹ç‹ç�WŽW�Ç�Ç’# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!17~!1è7~!1Õ¾!1dÕ¾!1ƒÅ!1lƒÅ!1 Ï!1è Ï!1CÒ!1´CÒ!1 è!1@ è!1§("1¼§("1U/"1ÄU/"1ß8"1@ß8"1<"1 <"1ÛQ"1˜ÛQ"1y’"1y’"1'™"1 '™"1±¢"1˜ ±¢"1ç¥"1d ç¥"1)Ó"1ð )Ó"1UÚ"1Ì UÚ"1µþ"1H µþ"1÷+#1h÷+#1;#1D;#1¿o#1пo#1s‘#1ôs‘#1qœ#1 qœ#1«$1¬«$1ƒò$1pƒò$1½Þ%1t½Þ%17ì%1T7ì%1i€&1ài€&19R'1$9R'1ád'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1®â�\ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“zºæ& bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1®â�\ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“zºæ& bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1®â�\  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“zºæ&  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ƒ£Ãã6V bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1{»6vï/E ¨è  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d{ûvöqñlìgçbâ]ÝXØQÑJ Ê C à œ  < ¼ { û 5 µ Z Ú . ® 9 ¹ '§  [õ5Ö�ÏHˆ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛOO��);*;4¯4¯>!?!I“I“ST^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››ƒƒ÷÷k k"ß"ß$ % '”'”)*,|,|.ð.ð0›1›346ƒ6ƒ8÷8÷:k;k=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b™b™d e g}g}iïiïkalanÓnÓpEqEs·s·u)v)x›x›z { }}ññ�c‚c„Õ„Õ†G‡G‰¸‰¸‹)Œ)ŽšŽš� ‘ “  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1›}!1è›}!19¾!1d9¾!1ƒÅ!1lƒÅ!1 Ï!1è Ï!1CÒ!1´CÒ!1mç!1@mç!1 ("1¼ ("1U/"1ÄU/"1ß8"1@ß8"1<"1 <"1?Q"1˜?Q"1Ý‘"1Ý‘"1'™"1 '™"1±¢"1˜ ±¢"1ç¥"1d ç¥"1)Ó"1ð )Ó"1yÚ"1Ì yÚ"1Ùþ"1H Ùþ"1,#1h,#1�:#1D�:#1Yo#1ÐYo#1W�#1ôW�#1Uš#1 Uš#1½$1¬½$1•ò$1p•ò$1ÏÞ%1tÏÞ%1Iì%1TIì%1É€&1àÉ€&1™R'1$™R'1Ae'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ò¾ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 HS“æ& bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ò¾ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 HS“æ& bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ò¾  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 HS“æ&  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‰©Rr,L bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1^žü<+kœÜ<|  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d^Þ<¼šøxÖV´4’pðmíPÐ8¸0°ƒ � ÎN ð p Ð P ° 0 �  p ð Û NŽì,ŒÌ,l bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ø ØMMÂÂ)7*74¬4¬>"?"I`I`ST^ˆ^ˆhühürrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ØØPPÈÈ @ @ ¸ ¸00¨¨  ˜˜  ‚‚÷÷l l"á"á$V%V'Ë'Ë)@*@,µ,µ.*/*1Ÿ1Ÿ346‰6‰8þ8þ:s;s=è=è?]@]BÒBÒDGEGG¼G¼I2J2L¦L¦NOQŽQŽSTV@V@X´X´Z([(]œ]œ_`bÄbÄd8e8g¬g¬i j l”l”noq|q|sðsðudvdxØxØzL{L}À}À4€4‚¨‚¨„…‡’‡’‰ŠŒ|Œ|ŽñŽñ�f‘f“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1+~!1è+~!1ɾ!1dɾ!1ƒÅ!1lƒÅ!1 Ï!1è Ï!1CÒ!1´CÒ!1ýç!1@ýç!1›("1¼›("1U/"1ÄU/"1ß8"1@ß8"1<"1 <"1ÏQ"1˜ÏQ"1m’"1m’"1'™"1 '™"1±¢"1˜ ±¢"1ç¥"1d ç¥"1)Ó"1ð )Ó"1£Ú"1Ì £Ú"1ÿ"1H ÿ"1E,#1hE,#1:#1D:#1;o#1Ð;o#1ÍŽ#1ôÍŽ#1Ë™#1 Ë™#1Ï$1¬Ï$1§ò$1p§ò$1áÞ%1táÞ%1[ì%1T[ì%1)�&1à)�&1ùR'1$ùR'1¡e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¡Õ�M b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“¥åæ& bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¡Õ�M b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“¥åæ& bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¡Õ�M  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“¥åæ&  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w— -y™ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1W—î.‡ÇU• `  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dW×.®…Ü\³3Š aá8¸‘ƒêjà`ÃC ½= œ  š  u õ N Î ' §  € × n®G%e à bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Þ ÞWWÐÐ)I*I4Â4Â>;?;I²I²S'T'^f^fhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÞÞWWÐÐ I I  Â;;´´--¦¦˜˜ Š Š"#%|%|'õ'õ)n*n,ç,ç.`/`1Ù1Ù3R4R6Ë6Ë8D9D;½;½=6>6@¯@¯B(C(E¡E¡GHJ“J“L M O�O�QøQøSoToVæVæX]Y][Ô[Ô]K^K`Â`Âb7c7e¬e¬g!h!j–j–l m oJoJq½q½s0t0v£v£xy{¿{¿}2~2€¥€¥‚ƒ…‹…‹‡þ‡þ‰pŠpŒâŒâŽT�T‘ƑƓ  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1›}!1è›}!19¾!1d9¾!1ƒÅ!1lƒÅ!1 Ï!1è Ï!1CÒ!1´CÒ!1mç!1@mç!1 ("1¼ ("1U/"1ÄU/"1ß8"1@ß8"1<"1 <"1?Q"1˜?Q"1Ý‘"1Ý‘"1'™"1 '™"1±¢"1˜ ±¢"1ç¥"1d ç¥"1)Ó"1ð )Ó"1OÚ"1Ì OÚ"1¯þ"1H ¯þ"1ñ+#1hñ+#1y:#1Dy:#15o#1Ð5o#1£Ž#1ô£Ž#1¡™#1 ¡™#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1�€&1à�€&1QR'1$QR'1ùd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ð¼ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 GS“æ& bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ð¼ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 GS“æ& bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡ð¼  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 GS“æ&  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 s“%Eö bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1W—î.ƒÃ¶ö X  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dW×.®…Ü\³3Š aá8¸ �âb·7 Œ ™  a á d ä 6 ¶ / ¯  ‹ ú z à ` · ÷ NŽf¦ã#x¸ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ö ÖKKÀÀ)5*54ª4ª>?I[I[S T ^~^~hðhðrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††ûûppååZ Z"Ï"Ï$D%D'¹'¹).*.,£,£./1�1�346w6w8ì8ì:a;a=Ö=Ö?K@KBÀBÀD5E5GªGªIJL’L’NOQxQxSëSëU'V'XšXšZ [ ]€]€_ó_óa¤b¤degˆgˆiúiúklllnÞnÞpPqPsÂsÂu4v4x¦x¦z{}Š}Šüü�n‚n„à„à†T‡T‰ljNj:Œ:Ž­Ž­� ‘ “  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!11~!1è1~!1Ͼ!1dϾ!1ƒÅ!1lƒÅ!1 Ï!1è Ï!1CÒ!1´CÒ!1è!1@è!1¡("1¼¡("1U/"1ÄU/"1ß8"1@ß8"1<"1 <"1ÕQ"1˜ÕQ"1s’"1s’"1'™"1 '™"1±¢"1˜ ±¢"1ç¥"1d ç¥"1)Ó"1ð )Ó"1aÚ"1Ì aÚ"1Áþ"1H Áþ"1,#1h,#1m:#1Dm:#1)o#1Ð)o#1ëŽ#1ôëŽ#1é™#1 é™#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1·€&1à·€&1‡R'1$‡R'1/e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1£×…Q b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“¤äæ& bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1£×…Q b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“¤äæ& bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1£×…Q  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“¤äæ&  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 †¦Ãã<\ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1S“æ&y¹ L ò2  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dSÓ&¦ùyÌLŸròEŘëk¾>‘ d ä 7 · w ÷  Š T Ô Ý ] 1 ± ° 0  Ž ƒ à V©é<| bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>-?-I¢I¢ST^S^Shûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbåbådXeXgËgËi>j>l±l±nïnïp`q`sÑsÑuBvBx³x³z[{[}Ì}Ì=€=‚®‚®„…‡‘‡‘‰ŠŒsŒsŽäŽä�U‘U“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1¡}!1С}!1?¾!1L?¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1sç!1(sç!1("1¤("1[/"1¬[/"1å8"1(å8"1<"1ô<"1EQ"1€EQ"1ã‘"1üã‘"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1�Ú"1´ �Ú"1ýþ"10 ýþ"1?,#1P?,#1‹7#1,‹7#1#�#1˜#�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1ç€&1Ìç€&1·R'1·R'1_e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆó¿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TS“æ& bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆó¿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TS“æ& bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆó¿  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TS“æ&  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ž®o�2R bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1(h€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1vöCÃ�Ý] ø 8 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` × ×NNÅÅ)<*<4³4³>,?,I£I£ST^Œ^Œhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``××NNÅÅ < < ³ ³**¡¡��}}ôôk k"â"â$Y%Y'Ð'Ð)G*G,¾,¾.5/51¬1¬3#4#6š6š89;ˆ;ˆ=ÿ=ÿ?v@vBíBíDdEdGÛGÛITJTLËLËNBOBQ¹Q¹S0T0V§V§XY[•[•] ^ `ƒ`ƒbúbúdqeqgègèi_j_lÖlÖnHoHqºqºs,t,vžvžxy{‚{‚}ô}ôf€f‚؂؄J…J‡¼‡¼‰.Š.Œ Œ Ž�‘„‘„“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1%~!1Ð%~!1þ!1Lþ!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1÷ç!1(÷ç!1•("1¤•("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1ÉQ"1€ÉQ"1g’"1üg’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1ÁÚ"1´ ÁÚ"1!ÿ"10 !ÿ"1c,#1Pc,#1£7#1,£7#1Ž#1˜Ž#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1;�&1Ì;�&1 S'1 S'1³e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–ÊvB b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“ƒÃê* bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–ÊvB b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“ƒÃê* bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–ÊvB  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“ƒÃê*  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 v–/–¶ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D„€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1æfÁAœ w ÷ ô4 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝUUÍÍ)E*E4½4½>5?5I«I«S!T!^—^—h i sƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%���� } }"õ"õ$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=>@•@•B C E…E…GýGýIuJuLëLëNaOaQ×Q×SMTMVÃVÃX9Y9[¯[¯]%^%`›`›bce‡e‡gýgýisjslélén_o_qÕqÕsKtKvÁvÁx7y7{­{­}#~#€™€™‚ƒ…………‡û‡û‰qŠqŒçŒçŽ]�]‘Ó‘Ó“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1¡}!1С}!1?¾!1L?¾!1‰Å!1T‰Å!1Ï!1ÐÏ!1IÒ!1œIÒ!1sç!1(sç!1("1¤("1[/"1¬[/"1å8"1(å8"1<"1ô<"1EQ"1€EQ"1ã‘"1üã‘"1-™"1 -™"1·¢"1€ ·¢"1í¥"1L í¥"1/Ó"1Ø /Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1:#1,:#1³�#1˜³�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆñ½ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 US“æ& bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆñ½ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 US“æ& bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆñ½  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 US“æ&  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 t”Jjù bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1µõ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ª * w ÷ D Ä  ‘ …Å €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Õ ÕJJ¿¿)4*44©4©> ? I•I•S T ^|^|hîhîr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÕÕJJ¿¿ 4 4 © ©““}}òòggÜÜQ Q"Æ"Æ$;%;'°'°)%*%,š,š./1„1„3ù3ù5n6n8ã8ã:X;X=Í=Í?B@BB·B·D,E,G¡G¡IJL�L�NOQwQwSìSìUaVaXÖXÖZK[K]À]À_5`5bªbªdeg”g”i j l~l~nðnðpbqbsÔsÔuFvFx¸x¸z*{*}œ}œ€‚€‚€„ò„ò†d‡d‰Ö‰Ö‹HŒHŽºŽº�,‘,“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1+~!1Ð+~!1ɾ!1Lɾ!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1÷ç!1(÷ç!1•("1¤•("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1ÉQ"1€ÉQ"1g’"1üg’"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1mÚ"1´ mÚ"1Íþ"10 Íþ"1,#1P,#1=:#1,=:#1»Ž#1˜»Ž#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜ÌzF b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“ƒÃë+ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜ÌzF b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“ƒÃë+ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜ÌzF  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“ƒÃë+  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ã㉩Bb bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ß €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1R Ò - ­  ˆ ã c �Ï €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙOOÅÅ);*;4±4±>'?'I›I›ST^ƒ^ƒh÷h÷rksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰ÿÿuuëëa a"×"×$M%M'Ã'Ã)9*9,¯,¯.%/%1›1›346‡6‡8ý8ý:s;s=é=é?_@_BÕBÕDKEKGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡‹‡‹‰ÿ‰ÿ‹sŒsŽçŽç�[‘[“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1§}!1Ч}!1E¾!1LE¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1yç!1(yç!1("1¤("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1KQ"1€KQ"1é‘"1üé‘"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#1…7#1,…7#1A�#1˜A�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ÿ€&1Ìÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ôÀ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @U•è( bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ôÀ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @U•è( bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ôÀ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @U•è(  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 •µt”:Z bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1#c€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1}ýLÌ›êj S €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Þ ÞXXÎÎ)D*D4º4º>0?0I¤I¤ST^Œ^Œhistst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÞÞXXÒÒ L L Æ Æ@@ºº44®®((¢¢ – –"#%†%†'ü'ü)r*r,è,è.^/^1Ô1Ô3J4J6À6À8696;¬;¬=">"@˜@˜BCE„E„GúGúIpJpLäLäNXOXQÌQÌS@T@V´V´X(Y([œ[œ]^`„`„bøbødlelgàgàiTjTlÈlÈn/?/I¤I¤ST^Ž^Žhisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbébéd^e^gÓgÓiHjHl½l½n2o2q§q§stv‘v‘xy{{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1§}!1Ч}!1E¾!1LE¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1sç!1(sç!1("1¤("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1EQ"1€EQ"1ã‘"1üã‘"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1:#1,:#1Ñ�#1˜Ñ�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ò¾ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @T”ç' bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ò¾ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @T”ç' bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ò¾  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @T”ç'  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u•ÿ]} bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1²ò €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¹ 9 ˆ  W × & ¦ ¢â €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜTTÈÈ)<*<4°4°>$?$I–I–ST^z^zhìhìr^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœŒŒ | |"ô"ô$h%h'Ü'Ü)P*P,Ä,Ä.8/81¬1¬3 4 6”6”89;|;|=ð=ð?d@dBØBØDLELGÀGÀI4J4L¦L¦NOQŠQŠSüSüUnVnXàXàZR[R]Ä]Ä_6`6b¨b¨degŒgŒiþiþkplpnânâpTqTsÆsÆu8v8xªxªz{}Ž}Ž€‚r‚r„ä„ä†V‡V‰ȉȋ:Œ:ެެ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1%~!1Ð%~!1þ!1Lþ!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1ñç!1(ñç!1�("1¤�("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1ÃQ"1€ÃQ"1a’"1üa’"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#11:#1,1:#1—Ž#1˜—Ž#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Áo; b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“™Ùö6 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Áo; b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“™Ùö6 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Áo;  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“™Ùö6  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ã㌬Hh bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Î €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations17 ·  � é i  B h¨ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÅÅ):*:4¯4¯>$?$I—I—S T ^}^}hðhðrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=ß=ß?T@TBÉBÉD>E>G³G³I(J(L›L›NOQ�Q�SôSôUgVgXÚXÚZM[M]À]À_3`3b¦b¦degŒgŒiÿiÿkrlrnånåpXqXsËsËu>v>x±x±z${$}—}— € ‚}‚}„ð„ð†c‡c‰Ö‰Ö‹IŒI޼޼�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1­}!1Э}!1K¾!1LK¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1ç!1(ç!1("1¤("1/"1¬/"1 9"1( 9"1?<"1ô?<"1QQ"1€QQ"1ï‘"1üï‘"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1mÚ"1´ mÚ"1Íþ"10 Íþ"1,#1P,#1‘7#1,‘7#1›�#1˜›�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠõÁ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Ik«@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠõÁ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Ik«@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠõÁ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Ik«@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ãã–¶<\ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1&f€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1˜míB — f¦ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×LLÁÁ)6*64«4«>?I‘I‘ST^w^whêhêr]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡üüqqææ[ ["Ð"Ð$E%E'º'º)/*/,¤,¤./1Ž1Ž346x6x8í8í:b;b=×=×?J@JB½B½D0E0G£G£IJL‰L‰NüNüPoQoSâSâUUVUXÈXÈZ;[;]®]®_!`!b”b”degzgziíiík`l`nÓnÓpFqFs¹s¹u,v,xŸxŸz{}…}…øø�k‚k„Þ„Þ†Q‡Q‰ÉË5Œ5ާާ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1~!1Ð~!1·¾!1L·¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1åç!1(åç!1ƒ("1¤ƒ("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1·Q"1€·Q"1U’"1üU’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1=Ú"1´ =Ú"1�þ"10 �þ"1ß+#1Pß+#1µ7#1,µ7#1ÁŽ#1˜ÁŽ#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ç€&1Ìç€&1·R'1·R'1_e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´`, b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“ƒÃé) bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´`, b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“ƒÃé) bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´`,  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“ƒÃé)  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p�Ää/ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1JŠ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1(¨ ‹în Ñ Q ŠÊ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii ã ãYYÏÏ)E*E4»4»>/?/I£I£ST^‹^‹hÿhÿrsss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiãã]]×× Q Q Ë ËEE¿¿99³³))ŸŸ ‹ ‹"#%w%w'í'í)c*c,Ù,Ù.O/O1Å1Å3;4;6±6±8'9';�;�=>@‡@‡BûBûDoEoGãGãIWJWLËLËN?O?Q³Q³S'T'V›V›XY[ƒ[ƒ]÷]÷_k`kbßbßdSeSgÇgÇi;j;l¯l¯n#o#q—q—s t vvxóxózg{g}Û}ÛO€O‚ÂÄ7…7‡«‡«‰ŠŒ‘Œ‘Ž�‘w‘w“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1­}!1Э}!1K¾!1LK¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1yç!1(yç!1("1¤("1y/"1¬y/"19"1(9"19<"1ô9<"1KQ"1€KQ"1é‘"1üé‘"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"17Ú"1´ 7Ú"1—þ"10 —þ"1Ù+#1PÙ+#1%:#1,%:#11Ž#1˜1Ž#1Ÿ$1$Ÿ$1wò$1èwò$1±Þ%1ì±Þ%1'€&1Ì'€&1÷Q'1÷Q'1Ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠó¿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jjªþ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠó¿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jjªþ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠó¿  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jjªþ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ããv–% bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1»û €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ì l Á A –  k ë û; €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙLL¿¿)2*24¥4¥>?I‡I‡SøSø]i^ihÚhÚrKsK} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰üüooââU U"È"È$;%;'®'®)!*!,”,”./1z1z3í3í5`6`8Ó8Ó:F;F=¹=¹?*@*B›B›D E G}G}IîIîK_L_NÐNÐPAQAS²S²U#V#X”X”Z[]v]v_ç_çaXbXdÉdÉf:g:i«i«kln�n�pþpþrosouàuàwQxQzÂzÂ|3}3¤¤�‚„†„††÷†÷ˆg‰g‹׋×�GŽG�·�·’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1~!1Ð~!1·¾!1L·¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ëç!1(ëç!1‰("1¤‰("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1½Q"1€½Q"1[’"1ü[’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1U:#1,U:#1c�#1˜c�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1o€&1Ìo€&1?R'1?R'1çd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶d0 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“ƒÃê* bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶d0 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“ƒÃê* bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶d0  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“ƒÃê*  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �¯ÃãNn bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1í- €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1´ 4 —  z ú ] Ý -m €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßSSÇÇ);*;4¯4¯>!?!I“I“ST^w^whéhér[s[} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ‡‡ûûo o"ã"ã$W%W'Ë'Ë)?*?,³,³.'/'1›1›346ƒ6ƒ8÷8÷:k;k=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b�b�dÿdÿfqgqiãiãkUlUnÇnÇp9q9s«s«uvx�x�z{}s}såå�W‚W„ɄɆ;‡;‰¬‰¬‹ŒŽŽŽŽ�ÿ�ÿ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1³}!1г}!1Q¾!1LQ¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ç!1(ç!1("1¤("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1QQ"1€QQ"1ï‘"1üï‘"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#1©7#1,©7#1gŽ#1˜gŽ#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @T”ç' bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @T”ç' bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ö  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @T”ç'  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ãã—·>^ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1<|€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ðpÐP°0 �  ,l €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜQQÆÆ);*;4°4°>&?&IšIšST^‚^‚höhörlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœ††ûûp p"å"å$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ì=ì?a@aBÖBÖDKEKGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqvqvsêsêu^v^xÒxÒzF{F}º}º.€.‚¢‚¢„…‡Œ‡Œ‰ŠŒvŒvŽëŽë�`‘`“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1~!1Ð~!1±¾!1L±¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ßç!1(ßç!1}("1¤}("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1±Q"1€±Q"1O’"1üO’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"11Ú"1´ 1Ú"1‘þ"10 ‘þ"1Ó+#1PÓ+#1Í7#1,Í7#1‡�#1˜‡�#1á$1$á$1¹ò$1è¹ò$1óÞ%1ìóÞ%1;�&1Ì;�&1 S'1 S'1³e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©U! b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S““Óç' bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©U! b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S““Óç' bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1u©U!  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S““Óç'  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ããzšý bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1` €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€nî\ Ü J Ê P� €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii â â[[ÔÔ)M*M4Æ4Æ>???I¶I¶S+T+^ž^žhis„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiââ[[ÔÔ M M Æ Æ??¸¸11ªª##œœ Ž Ž"#%€%€'ù'ù)r*r,ë,ë.d/d1Ý1Ý3V4V6Ï6Ï8H9H;Á;Á=:>:@³@³B,C,E¥E¥GHJ—J—LMO…O…QüQüSsTsVêVêXaYa[Ø[Ø]O^O`Æ`Æb;c;e°e°g%h%jšjšlmo‚o‚qõqõshthvÛvÛxNyN{Á{Á}4~4€§€§‚ƒ…�…�‡ˆŠrŠrŒäŒäŽV�V‘ȑȓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1­}!1Э}!1K¾!1LK¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ç!1(ç!1("1¤("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1QQ"1€QQ"1ï‘"1üï‘"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1=Ú"1´ =Ú"1�þ"10 �þ"1ß+#1Pß+#1I:#1,I:#1�#1˜�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ôÀ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @T”ç' bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ôÀ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @T”ç' bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ôÀ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @T”ç'  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ããw— ) bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ü €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1p ð P Ð 0 °  � Ì  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94®4®>#?#I–I–ST^z^zhìhìr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?O@OBÄBÄD9E9G®G®I#J#L–L–N O Q|Q|SïSïUbVbXÕXÕZH[H]»]»_.`.b b deg„g„iöiökhlhnÚnÚpLqLs¾s¾u0v0x¢x¢z{}†}†øø�j‚j„܄܆P‡P‰ÉË6Œ6ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1~!1Ð~!1±¾!1L±¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1åç!1(åç!1ƒ("1¤ƒ("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1·Q"1€·Q"1U’"1üU’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1yÚ"1´ yÚ"1Ùþ"10 Ùþ"1,#1P,#1y:#1,y:#15�#1˜5�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«Y% b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S““Óç' bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«Y% b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S““Óç' bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«Y%  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S““Óç'  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ã㊪Dd bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1N €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18 ¸ & ¦  ” ‚þ> €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Þ ÞUUÌÌ)C*C4º4º>1?1I¦I¦ST^Š^Šhûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIYJYLÎLÎNCOCQ¸Q¸S-T-V¢V¢XY[Œ[Œ]^`v`vbébéd\e\gÏgÏiBjBlµlµn&o&q—q—stvyvyxêxêz[{[}Ì}Ì=€=‚®‚®„…‡‘‡‘‰ŠŒsŒsŽäŽä�U‘U“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1¹}!1й}!1W¾!1LW¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1…ç!1(…ç!1#("1¤#("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1WQ"1€WQ"1õ‘"1üõ‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1Á7#1,Á7#13�#1˜3�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1ÿ€&1Ìÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ÷à b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @T”ç' bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ÷à b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @T”ç' bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ÷à  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @T”ç'  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ã㎮,L bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1R’€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1JÊ5µ   ‹ ô4 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>0?0I§I§ST^�^�histst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIXJXLÏLÏNFOFQ½Q½S4T4V«V«X"Y"[™[™]^`‡`‡bþbþdueugìgìicjclÚlÚnLoLq¾q¾s0t0v¢v¢xy{†{†}ø}øj€j‚܂܄N…N‡À‡À‰2Š2Œ¤Œ¤Ž�‘ˆ‘ˆ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1~!1Ð~!1¥¾!1L¥¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1Ùç!1(Ùç!1w("1¤w("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1«Q"1€«Q"1I’"1üI’"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1+Ú"1´ +Ú"1‹þ"10 ‹þ"1Í+#1PÍ+#1å7#1,å7#1S�#1˜S�#1ç$1$ç$1¿ò$1è¿ò$1ùÞ%1ìùÞ%1M�&1ÌM�&1S'1S'1Åe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžJ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“…Åæ& bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžJ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“…Åæ& bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžJ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“…Åæ&  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ããa�ò bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1v¶€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ØXÑQ Ê J à C V €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áYYÑÑ)I*I4Á4Á>9?9I¯I¯S%T%^›^›his‡s‡} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‘‘  � �"ù"ù$q%q'é'é)a*a,Ù,Ù.Q/Q1É1É3A4A6¹6¹8191;©;©=!>!@™@™BCE‰E‰GHJyJyLïLïNeOeQÛQÛSQTQVÇVÇX=Y=[³[³])^)`Ÿ`Ÿbce‹e‹ghjwjwlílíncocqÙqÙsOtOvÅvÅx;y;{±{±}'~'€�€�‚ƒ…‰…‰‡ÿ‡ÿ‰uŠuŒëŒëŽa�a‘בד b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1³}!1г}!1Q¾!1LQ¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1…ç!1(…ç!1#("1¤#("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1WQ"1€WQ"1õ‘"1üõ‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"11Ú"1´ 1Ú"1‘þ"10 ‘þ"1Ó+#1PÓ+#1m:#1,m:#1Û�#1˜Û�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1¥€&1Ì¥€&1uR'1uR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒõÁ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @T”ç' bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒõÁ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @T”ç' bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒõÁ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @T”ç'  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÃãnŽù bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ý= €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ö v á a Ì L · 7Ÿß €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ù ÙNNÃÃ)8*84­4­>$?$I™I™ST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÙÙNNÃà 8 8 ­ ­""——  ��öökkààU U"Ê"Ê$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1 ~!1Ð ~!1«¾!1L«¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1ßç!1(ßç!1}("1¤}("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1±Q"1€±Q"1O’"1üO’"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1�:#1,�:#1 ‘#1˜ ‘#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1Õ€&1ÌÕ€&1¥R'1¥R'1Me'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l N b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S“„Äæ& bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l N b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S“„Äæ& bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l N  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S“„Äæ&  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ãã�¡.N bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1/o €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¼ < µ 5 ® .§'Ï €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝSSÉÉ)?*?4µ4µ>+?+IŸIŸST^‡^‡hûhûroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��yyïïe e"Û"Û$Q%Q'Ç'Ç)=*=,³,³.)/)1Ÿ1Ÿ346‹6‹89;w;w=í=í?c@cBÙBÙDOEOGÅGÅI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÃbÃd7e7g«g«ijl“l“noq{q{sïsïucvcx×x×zK{K}¿}¿3€3‚§‚§„…‡�‡�‰ŠŒwŒwŽëŽë�_‘_“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1¹}!1й}!1W¾!1LW¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1‹ç!1(‹ç!1)("1¤)("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1]Q"1€]Q"1û‘"1üû‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1ß7#1,ß7#1ù�#1˜ù�#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1�&1Ì�&1áR'1áR'1‰e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�øÄ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @T”ç' bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�øÄ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @T”ç' bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�øÄ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @T”ç'  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ãャ6 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1h¨€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  – Œ ‚  ¸ø €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh â â\\ÒÒ)H*H4¾4¾>4?4I¨I¨ST^�^�hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²,,¦¦  š š"#%Š%Š'(*v*v,ì,ì.b/b1Ø1Ø3N4N6Ä6Ä8:9:;°;°=&>&@œ@œBCEˆEˆGþGþItJtLèLèN\O\QÐQÐSDTDV¸V¸X,Y,[ [ ]^`ˆ`ˆbübüdpepgägäiXjXlÌlÌn@o@q´q´s(t(vœvœxy{„{„}ø}øl€l‚à‚à„T…T‡ȇȉ<Š<Œ°Œ°Ž$�$‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1~!1Ð~!1Ÿ¾!1LŸ¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1Óç!1(Óç!1q("1¤q("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1¥Q"1€¥Q"1C’"1üC’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1Ú"1´ Ú"1þ"10 þ"1Á+#1PÁ+#18#1,8#1‘#1˜‘#1á$1$á$1¹ò$1è¹ò$1óÞ%1ìóÞ%1�&1Ì�&1çR'1çR'1�e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“?  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S““Óç' bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“?  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S““Óç' bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“?   b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S““Óç'  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÃãXxà bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ŒÌ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations10°4 ´ 8 ¸ < ¼ Ü €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii à àWWÎÎ)E*E4¼4¼>3?3I¨I¨ST^’^’his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiààWWÎÎ E E ¼ ¼33ªª!!˜˜††ýýt t"ë"ë$b%b'Ù'Ù)P*P,Ç,Ç.>/>1µ1µ3,4,6£6£89;‘;‘=>@@BöBöDmEmGäGäI[J[LÐLÐNEOEQºQºS/T/V¤V¤XY[Ž[Ž]^`x`xbíbídbebg×g×iLjLlÁlÁn6o6q«q«s t v•v•x y {{}ô}ôi€i‚Þ‚Þ„S…S‡ȇȉ=Š=Œ²Œ²Ž'�'‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1¹}!1й}!1W¾!1LW¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1‹ç!1(‹ç!1)("1¤)("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1]Q"1€]Q"1û‘"1üû‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1%Ú"1´ %Ú"1…þ"10 …þ"1Ç+#1PÇ+#1‘:#1,‘:#1­�#1˜­�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @T”ç' bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @T”ç' bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�ö  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @T”ç'  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ããcƒç bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1^ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1x ø n î d ä ZÚn® €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àXXÌÌ)@*@4´4´>(?(IšIšS T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  �� € €"ø"ø$l%l'à'à)T*T,È,È.Œ>ްް�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1~!1Ð~!1¥¾!1L¥¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1Ùç!1(Ùç!1w("1¤w("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1«Q"1€«Q"1I’"1üI’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#1Ç:#1,Ç:#1ß‘#1˜ß‘#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1«€&1Ì«€&1{R'1{R'1#e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•C b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 S““Óç' bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•C b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 S““Óç' bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•C  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 S““Óç'  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ããx˜8 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1P� €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@ À D Ä HÈLÌ à €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞTTÉÉ)>*>4³4³>(?(I›I›ST^�^�hôhôrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽzzððf f"Ü"Ü$Q%Q'Æ'Æ);*;,°,°.%/%1š1š346„6„8ù8ù:n;n=ã=ã?X@XBÍBÍDBEBG·G·I,J,LŸLŸNOQ…Q…SøSøUkVkXÞXÞZQ[Q]Ä]Ä_7`7bªbªdeg�g�ijlvlvnénép\q\sÏsÏuBvBxµxµz({(}›}›€‚�‚�„ô„ô†g‡g‰Ú‰Ú‹MŒMŽÀŽÀ�3‘3“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1×}!1�×}!1u¾!1u¾!1¹Â!1$¹Â!1£ç!1�£ç!1A("1A("1‹,"1$‹,"1uQ"1�uQ"1’"1’"1]–"1$]–"1¯Ú"1�¯Ú"1ÿ"1üÿ"150#1 50#1#1| #1Û$1Ü Û$1³ò$1  ³ò$1íÞ%1¤ íÞ%1«€&1„ «€&1{R'1È{R'1#e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 =kŸ ÖÙ¥ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 =kŸ ÖÙ¥ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 =kŸ ÖÙ¥ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1d„  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ì ìaaÖÖ)K*K4¾4¾>1?1I¤I¤ST^Š^Šhýhýrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvììbbØØ N N Ä Ä::°°&&œœ††ûûp p"å"å$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ê=ê?]@]BÐBÐDCECG¶G¶I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnænæpYqYsÌsÌu?v?x²x²z%{%}˜}˜ € ‚~‚~„ñ„ñ†b‡b‰Ô‰Ô‹FŒFޏޏ�*‘*“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1Ÿ¾!1 Ÿ¾!1SÅ!1SÅ!1Íç!1€Íç!1k("1ük("1%/"1%/"1ŸQ"1pŸQ"1=’"1ì=’"1÷˜"1ô÷˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"150#1ì50#1#1L #1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡3ÿ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1›Û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡3ÿ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1›Û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡3ÿ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1›Û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1}�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ð ðffÜÜ)R*R4Æ4Æ>:?:I®I®S"T"^–^–h i s|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvððjjää ^ ^ Ø ØRRÌÌFFÀÀ66¬¬"" ˜ ˜"#%„%„'ú'ú)p*p,æ,æ.\/\1Ò1Ò3H4H6¾6¾8494;ª;ª=>@’@’BCEzEzGîGîIbJbLÖLÖNJOJQ¾Q¾S2T2V¦V¦XY[Ž[Ž]^`v`vbêbêd^e^gÒgÒiFjFlºlºn.o.q¢q¢stvŠvŠxþxþzr{r}æ}æZ€Z‚΂΄B…B‡´‡´‰'Š'ŒšŒšŽ � ‘€‘€“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Ñ}!1�Ñ}!1o¾!1o¾!1¹Â!1$¹Â!1£ç!1�£ç!1A("1A("1‹,"1$‹,"1uQ"1�uQ"1’"1’"1]–"1$]–"1¯Ú"1�¯Ú"1ÿ"1üÿ"150#1 50#1#1| #1Ã$1Ü Ã$1›ò$1  ›ò$1ÕÞ%1¤ ÕÞ%1E€&1„ E€&1R'1ÈR'1½d'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 >i�Òܨ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 >i�Òܨ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 >i�Òܨ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1g‡  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ì ì__ÒÒ)E*E4¶4¶>'?'I˜I˜S T ^z^zhëhërZsZ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvììbbØØ N N Ä Ä::°°&&œœ‚‚õõh h"Û"Û$N%N'Á'Á)4*4,§,§./1�1�346s6s8æ8æ:Y;Y=Ê=Ê?;@;B¬B¬DEGŽGŽIÿIÿKpLpNáNáPRQRSÃSÃU4V4X¥X¥Z[]‡]‡_ø_øaibidÚdÚfKgKi¼i¼k-l-nžnžpqs€s€uñuñwbxbzÓzÓ|D}Dµµ�&‚&„—„—†‡‰v‰v‹æ‹æ�VŽV�Æ�Æ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1Ÿ¾!1 Ÿ¾!1SÅ!1SÅ!1Óç!1€Óç!1q("1üq("1%/"1%/"1¥Q"1p¥Q"1C’"1ìC’"1÷˜"1ô÷˜"1ÓÚ"1`ÓÚ"13ÿ"1Ì3ÿ"150#1ì50#1#1L #1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡5 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1šÚ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡5 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1šÚ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1S‡5 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1šÚ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ˆ¨  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv î îbbÖÖ)J*J4¼4¼>.?.I I ST^„^„höhörfsf} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvîîffÞÞ V V Î ÎFF¾¾66®®""––  ~ ~"ò"ò$f%f'Ú'Ú)N*N,Â,Â.6/61ª1ª346’6’89;z;z=ì=ì?^@^BÐBÐDBEBG´G´I&J&L˜L˜N O Q|Q|SîSîU`V`XÒXÒZD[D]¶]¶_(`(bšbšd e g~g~iðiðkblbnÔnÔpFqFs¸s¸u*v*xœxœz{}€}€òò�d‚d„Ö„Ö†F‡F‰·‰·‹(Œ(ޙޙ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Ñ}!1�Ñ}!1o¾!1 o¾!1¹Â!1¹Â!1�ç!1€�ç!1;("1ü;("1‹,"1‹,"1oQ"1poQ"1 ’"1ì ’"1]–"1ô]–"1£Ú"1`£Ú"1ÿ"1Ìÿ"150#1ì50#1#1L #1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1í€&1T í€&1½R'1˜½R'1ee'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœÓ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœÓ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœÓ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1]}  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu í íbb××)L*L4Á4Á>5?5I©I©ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuííeeÝÝ U U Í ÍEE½½55­­""——  � �"ö"ö$k%k'à'à)U*U,Ê,Ê.?/?1´1´3)4)6ž6ž89;ˆ;ˆ=ý=ý?r@rBçBçD\E\GÑGÑIEJEL¹L¹N-O-Q¡Q¡STV‰V‰XýXýZq[q]å]å_Y`YbÍbÍdAeAgµgµi)j)l�l�noq…q…sùsùumvmxáxázU{U}É}É=€=‚±‚±„%…%‡™‡™‰ŠŒƒŒƒŽøŽø�m‘m“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1Ÿ¾!1 Ÿ¾!1GÅ!1GÅ!1Óç!1€Óç!1q("1üq("1/"1/"1¥Q"1p¥Q"1C’"1ìC’"1ë˜"1ôë˜"1»Ú"1`»Ú"1ÿ"1Ìÿ"150#1ì50#1#1L #1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1M�&1T M�&1S'1˜S'1Åe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�< b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1“Ó€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�< b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1“Ó€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�< b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1“Ó €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1x˜  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ë ëddÝÝ)V*V4Ï4Ï>F?FI½I½S2T2^¥^¥his‰s‰} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrëëddÝÝ V V Ï ÏHHÁÁ::³³,,¥¥ — —"#%‰%‰'(*{*{,ô,ô.m/m1æ1æ3_4_6Ø6Ø8Q9Q;Ê;Ê=C>C@¼@¼B5C5E®E®G'H'JžJžLMOŒOŒQRTzTzVñVñXhYh[ß[ß]V^V`Í`ÍbBcBe·e·g,h,j¡j¡lmo‰o‰qüqüsotovâvâxUyU{È{È};~;€®€®‚!ƒ!…”…”‡ˆŠwŠwŒéŒéŽ[�[‘͓͑ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Ë}!1�Ë}!1i¾!1 i¾!1¹Â!1¹Â!1�ç!1€�ç!1;("1ü;("1‹,"1‹,"1oQ"1poQ"1 ’"1ì ’"1]–"1ô]–"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"150#1ì50#1#1L #1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÏ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÏ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÏ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1lŒ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss é é^^ÓÓ)H*H4½4½>0?0I£I£ST^‡^‡hùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðsséé__ÕÕ K K Á Á77­­##™™ƒƒøøm m"â"â$W%W'Ì'Ì)A*A,¶,¶.+/+1 1 346Š6Š8ÿ8ÿ:t;t=é=é?^@^BÓBÓDHEHG½G½I0J0L£L£NOQ‰Q‰SüSüUoVoXâXâZU[U]È]È_;`;b­b­deg‘g‘ijlulunçnçpYqYsËsËu=v=x¯x¯z!{!}“}“€‚w‚w„é„é†[‡[‰Ή΋AŒAŽ´Ž´�'‘'“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1¥¾!1 ¥¾!1GÅ!1GÅ!1Óç!1€Óç!1q("1üq("1/"1/"1¥Q"1p¥Q"1C’"1ìC’"1ë˜"1ôë˜"1ÍÚ"1`ÍÚ"1-ÿ"1Ì-ÿ"150#1ì50#1#1L #1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�>  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1“Ó€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�>  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1“Ó€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�>  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1“Ó €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1‡§  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr é é``××)N*N4Å4Å>:?:I¯I¯S"T"^“^“histst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrréé``×× N N Å Å<<³³**¡¡�� } }"ô"ô$k%k'â'â)Y*Y,Ð,Ð.G/G1¾1¾35456¬6¬8#9#;š;š=>@ˆ@ˆBÿBÿDvEvGíGíIbJbL×L×NLOLQÁQÁS6T6V«V«X Y [•[•] ^ ``bòbòdeeegØgØiKjKl¾l¾n/o/q q stv‚v‚xóxózd{d}Õ}ÕF€F‚·‚·„(…(‡˜‡˜‰ Š ŒzŒzŽëŽë�\‘\“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Å}!1�Å}!1c¾!1 c¾!1¿Â!1¿Â!1—ç!1€—ç!15("1ü5("1‘,"1‘,"1iQ"1piQ"1’"1ì’"1c–"1ôc–"1—Ú"1`—Ú"1÷þ"1Ì÷þ"1/0#1ì/0#1û~#1L û~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1 �&1T  �&1ÛR'1˜ÛR'1ƒe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ð b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ð b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ð b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Xx  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq è è__ÖÖ)M*M4Ä4Ä>;?;I²I²S)T)^›^›h i ss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqèè__ÖÖ M M Ä Ä;;²²))  ŽŽ | |"ó"ó$j%j'á'á)X*X,Ï,Ï.F/F1½1½34446«6«8"9";™;™=>@‡@‡BþBþDuEuGìGìIcJcLÚLÚNQOQQÈQÈS?T?V¶V¶X-Y-[¤[¤]^`’`’b c e€e€g÷g÷injnlålånWoWqÉqÉs;t;v­v­xy{‘{‘}~€u€u‚ç‚ç„Y…Y‡ˇˉ=Š=Œ¯Œ¯Ž!�!‘“‘““ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1¥¾!1 ¥¾!15Å!15Å!1Ùç!1€Ùç!1w("1üw("1/"1/"1«Q"1p«Q"1I’"1ìI’"1Ù˜"1ôÙ˜"1µÚ"1`µÚ"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1S�&1T S�&1#S'1˜#S'1Ëe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™E b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ƒÃ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™E b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ƒÃ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™E b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ƒÃ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1q‘  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn æ æ^^ÖÖ)N*N4Æ4Æ>>¶¶..¦¦–– † †"þ"þ$v%v'î'î)f*f,Þ,Þ.V/V1Î1Î3F4F6¾6¾8696;®;®=&>&@ž@žBCEŽEŽGHJ|J|LòLòNhOhQÞQÞSTTTVÊVÊX@Y@[¶[¶],^,`¢`¢bceŽeŽghjzjzlðlðnfofqÜqÜsRtRvÈvÈx>y>{´{´}*~*€ € ‚ƒ…Œ…Œ‡ˆŠxŠxŒîŒîŽd�d‘Ú‘Ú“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Å}!1�Å}!1c¾!1 c¾!1¿Â!1¿Â!1‘ç!1€‘ç!1/("1ü/("1‘,"1‘,"1cQ"1pcQ"1’"1ì’"1c–"1ôc–"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1±€&1T ±€&1�R'1˜�R'1)e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ì b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ì b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ì b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1k‹  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo ä äYYÎÎ)C*C4¸4¸>-?-I¢I¢ST^‰^‰hûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooääYYÎÎ C C ¸ ¸--¢¢ŒŒvvëë` `"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%LšLšNOQ„Q„SùSùUnVnXãXãZX[X]Í]Í_B`Bb·b·d,e,g¡g¡ijl‹l‹nýnýpoqosásáuSvSxÅxÅz7{7}©}©€‚�‚�„ÿ„ÿ†q‡q‰ã‰ã‹UŒUŽÇŽÇ�9‘9“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1 ~!1� ~!1«¾!1 «¾!15Å!15Å!1Ùç!1€Ùç!1w("1üw("1/"1/"1«Q"1p«Q"1I’"1ìI’"1Ù˜"1ôÙ˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1/0#1ì/0#1û~#1L û~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™G b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ƒÃ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™G b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ƒÃ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™G b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ƒÃ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ˆ¨  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn ä äZZÐÐ)F*F4¼4¼>0?0I¤I¤ST^Œ^Œhistst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnääZZÐÐ F F ¼ ¼22¨¨””  €€ööl l"â"â$X%X'Î'Î)D*D,º,º.0/01¦1¦346’6’89;~;~=ô=ô?j@jBàBàDVEVGÌGÌI@J@L´L´N(O(QœQœSTV„V„XøXøZl[l]à]à_T`TbÈbÈd;?;I¯I¯S#T#^—^—h i ss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqëëeeßß Y Y Ó ÓMMÇÇAA»»55¯¯)) £ £"#%“%“' ( **,õ,õ.k/k1á1á3W4W6Í6Í8C9C;¹;¹=/>/@¥@¥BCE‘E‘GHJ{J{LïLïNcOcQ×Q×SKTKV¿V¿X3Y3[§[§]^`�`�bcewewgëgëi_j_lÓlÓnGoGq»q»s/t/v£v£xy{‹{‹}ÿ}ÿs€s‚ç‚ç„[…[‡χωCŠCŒ·Œ·Ž+�+‘Ÿ‘Ÿ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1 ~!1� ~!1«¾!1 «¾!1MÅ!1MÅ!1ßç!1€ßç!1}("1ü}("1/"1/"1±Q"1p±Q"1O’"1ìO’"1ñ˜"1ôñ˜"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1)0#1ì)0#1õ~#1L õ~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢N b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1œÜ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢N b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1œÜ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢N b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1œÜ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1lŒ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj á áXXÏÏ)F*F4½4½>2?2I§I§ST^‘^‘his{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjááXXÏÏ F F ½ ½44««""™™‡‡þþu u"ì"ì$c%c'Ú'Ú)Q*Q,È,È.?/?1¶1¶3-4-6¤6¤89;’;’= > @€@€B÷B÷DnEnGåGåIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbìbìdaeagÖgÖiKjKlÀlÀn5o5qªqªstv”v”x y {~{~}ó}óh€h‚݂݄R…R‡LJlj<Š<Œ±Œ±Ž&�&‘›‘›“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1¿}!1�¿}!1]¾!1 ]¾!1­Â!1­Â!1‹ç!1€‹ç!1)("1ü)("1,"1,"1]Q"1p]Q"1û‘"1ìû‘"1Q–"1ôQ–"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”ýÉ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”ýÉ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”ýÉ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1lŒ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo ç ç__ÓÓ)G*G4»4»>-?-IŸIŸST^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooçç__×× O O Ç Ç??··//§§—— ‡ ‡"ÿ"ÿ$s%s'ç'ç)[*[,Ï,Ï.C/C1·1·3+4+6Ÿ6Ÿ89;‡;‡=û=û?o@oBãBãDWEWGËGËI=J=L¯L¯N!O!Q“Q“STVwVwXéXéZ[[[]Í]Í_?`?b±b±d#e#g•g•ijlylynënëp]q]sÏsÏuAvAx³x³z%{%}—}— € ‚{‚{„í„í†_‡_‰щыCŒC޵޵�'‘'“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1 ~!1� ~!1«¾!1 «¾!1MÅ!1MÅ!1ßç!1€ßç!1}("1ü}("1/"1/"1±Q"1p±Q"1O’"1ìO’"1ñ˜"1ôñ˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢P b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ý€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢P b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ý€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢P b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ý €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1‡§  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk á áWWÌÌ)A*A4¶4¶>)?)IœIœST^‚^‚hõhõrhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkááWWÍÍ C C ¹ ¹//¥¥‘‘}}óói i"ß"ß$T%T'É'É)>*>,³,³.(/(1�1�346‡6‡8ü8ü:q;q=æ=æ?[@[BÐBÐDEEEGºGºI-J-L L NOQ†Q†SùSùUlVlXßXßZR[R]Å]Å_8`8b«b«deg‘g‘ijlwlwnênêp]q]sÐsÐuCvCx¶x¶z){)}œ}œ€‚‚‚‚„õ„õ†h‡h‰Û‰Û‹NŒNŽÁŽÁ�4‘4“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1¹}!1�¹}!1W¾!1 W¾!1¹Â!1¹Â!1‹ç!1€‹ç!1)("1ü)("1‹,"1‹,"1]Q"1p]Q"1û‘"1ìû‘"1]–"1ô]–"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1)0#1ì)0#1õ~#1L õ~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“þÊ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“þÊ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“þÊ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ãã  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj à àUUÊÊ)?*?4²4²>%?%I˜I˜S T ^~^~hñhñrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjààVVÌÌ B B ¸ ¸..¤¤��zzïïd d"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=Þ=Þ?Q@QBÄBÄD7E7GªGªIJL�L�NOQvQvSéSéU\V\XÏXÏZB[B]µ]µ_(`(b›b›deg�g�iôiôkglgnÚnÚpMqMsÀsÀu3v3x¦x¦z{}Œ}Œÿÿ�r‚r„å„å†V‡V‰ȉȋ:Œ:ެެ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1~!1�~!1±¾!1 ±¾!1/Å!1/Å!1åç!1€åç!1ƒ("1üƒ("1/"1/"1·Q"1p·Q"1U’"1ìU’"1Ó˜"1ôÓ˜"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ç€&1T ç€&1·R'1˜·R'1_e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«W# b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1„Ä€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«W# b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1„Ä€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«W# b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1„Ä €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ãã  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ä äZZÐÐ)F*F4º4º>.?.I¢I¢ST^Š^Šhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjää^^ØØ R R Ì ÌFFÀÀ::´´**   Œ Œ"#%x%x'î'î)d*d,Ú,Ú.P/P1Æ1Æ3<4<6²6²8(9(;ž;ž=>@†@†BúBúDnEnGâGâIVJVLÊLÊN>O>Q²Q²S&T&VšVšXY[‚[‚]ö]ö_j`jbÞbÞdReRgÆgÆi:j:l®l®n"o"q–q–s t v~v~xòxòzf{f}Ú}ÚN€N‚‚„6…6‡¨‡¨‰ŠŒŽŒŽŽ�‘t‘t“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1³}!1�³}!1Q¾!1 Q¾!1¹Â!1¹Â!1…ç!1€…ç!1#("1ü#("1‹,"1‹,"1WQ"1pWQ"1õ‘"1ìõ‘"1]–"1ô]–"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1)0#1ì)0#1õ~#1L õ~#1«$1¬ «$1ƒò$1p ƒò$1½Þ%1t ½Þ%1-€&1T -€&1ýQ'1˜ýQ'1¥d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘úÆ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘úÆ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘úÆ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ãã  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj à àSSÆÆ)9*94ª4ª>?IŒIŒSýSý]n^nhßhßrNsN} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjààVVÌÌ B B ¸ ¸..¤¤��vvéé\ \"Ï"Ï$B%B'µ'µ)(*(,›,›./1�1�3ô3ô5g6g8Ú8Ú:M;M=¾=¾?/@/B B DEG‚G‚IóIóKdLdNÕNÕPFQFS·S·U(V(X™X™Z [ ]{]{_ì_ìa]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1~!1�~!1±¾!1 ±¾!1/Å!1/Å!1åç!1€åç!1ƒ("1üƒ("1/"1/"1·Q"1p·Q"1U’"1ìU’"1Ó˜"1ôÓ˜"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1)0#1ì)0#1õ~#1L õ~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1o€&1T o€&1?R'1˜?R'1çd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«Y% b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1„Ä€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«Y% b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1„Ä€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1w«Y% b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1„Ä €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ãã  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âVVÊÊ)>*>4°4°>"?"I”I”ST^x^xhêhêrZsZ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢ŠŠþþr r"æ"æ$Z%Z'Î'Î)B*B,¶,¶.*/*1ž1ž346†6†8ú8ú:n;n=à=à?R@RBÄBÄD6E6G¨G¨IJLŒLŒNþNþPpQpSâSâUTVTXÆXÆZ8[8]ª]ª_`bŽbŽdegrgriäiäkVlVnÈnÈp:q:s¬s¬uvx�x�z{}t}tææ�X‚X„ʄʆ:‡:‰«‰«‹ŒŽ�Ž��þ�þ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1³}!1�³}!1Q¾!1 Q¾!1§Â!1§Â!1…ç!1€…ç!1#("1ü#("1y,"1y,"1WQ"1pWQ"1õ‘"1ìõ‘"1K–"1ôK–"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1Ï€&1T Ï€&1ŸR'1˜ŸR'1Ge'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ûÇ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ûÇ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ûÇ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ãã  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áVVËË)@*@4µ4µ>)?)I�I�ST^…^…hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‹‹ u u"ê"ê$_%_'Ô'Ô)I*I,¾,¾.3/31¨1¨346’6’89;|;|=ñ=ñ?f@fBÛBÛDPEPGÅGÅI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒwŒwŽìŽì�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1~!1�~!1·¾!1 ·¾!1AÅ!1AÅ!1åç!1€åç!1ƒ("1üƒ("1/"1/"1·Q"1p·Q"1U’"1ìU’"1å˜"1ôå˜"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1/�&1T /�&1ÿR'1˜ÿR'1§e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´`, b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1˜Ø€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´`, b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1˜Ø€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1€´`, b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1˜Ø €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ãã  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff ß ßXXÑÑ)J*J4Ã4Ã>:?:I±I±S&T&^™^™h i s}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffßßXXÑÑ J J à Ã<<µµ..§§  ™™ ‹ ‹"#%}%}'ö'ö)o*o,è,è.a/a1Ú1Ú3S4S6Ì6Ì8E9E;¾;¾=7>7@°@°B)C)E¢E¢GHJ’J’L M O€O€Q÷Q÷SnTnVåVåX\Y\[Ó[Ó]J^J`Á`Áb6c6e«e«g h j•j•l m o}o}qðqðsctcvÖvÖxIyI{¼{¼}/~/€¢€¢‚ƒ…ˆ…ˆ‡ù‡ù‰kŠkŒ݌ݎO�O‘Á‘Á“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1­}!1�­}!1K¾!1 K¾!1§Â!1§Â!1ç!1€ç!1("1ü("1y,"1y,"1QQ"1pQQ"1ï‘"1ìï‘"1K–"1ôK–"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ÷à b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ÷à b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽ÷à b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ãã  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝRRÇÇ)<*<4±4±>$?$I—I—S T ^{^{híhír_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��wwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD.?.I£I£ST^‡^‡høhørhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÝÝTTËË B B ¹ ¹00§§••  ƒƒúúq q"è"è$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbæbædYeYgÌgÌi?j?l²l²n#o#q”q”stvvvvxçxçzX{X}É}É:€:‚«‚«„…‡Œ‡Œ‰ý‰ý‹nŒnŽߎß�P‘P“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1­}!1�­}!1K¾!1 K¾!1³Â!1³Â!1yç!1€yç!1("1ü("1…,"1…,"1KQ"1pKQ"1é‘"1ìé‘"1W–"1ôW–"1óØ"1`óØ"1Sý"1ÌSý"10#1ì0#1é~#1L é~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ó€&1T ó€&1ÃR'1˜ÃR'1ke'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�øÄ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�øÄ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�øÄ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ãã  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>/?/I¦I¦ST^‘^‘hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÎLÎNEOEQ¼Q¼S3T3VªVªX!Y![˜[˜]^`†`†býbýdtetgëgëibjblÙlÙnMoMq¿q¿s1t1v£v£xy{‡{‡}ù}ùk€k‚݂݄O…O‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1~!1�~!1½¾!1 ½¾!1#Å!1#Å!1ëç!1€ëç!1‰("1ü‰("1õ."1õ."1½Q"1p½Q"1[’"1ì[’"1ǘ"1ôǘ"1óØ"1`óØ"1Sý"1ÌSý"10#1ì0#1é~#1L é~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1A�&1T A�&1S'1˜S'1¹e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½i5 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ƒÃ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½i5 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ƒÃ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½i5 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ƒÃ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ãã  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚRRÊÊ)B*B4º4º>0?0I¦I¦ST^œ^œhisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""ššŠŠ z z"ò"ò$j%j'â'â)Z*Z,Ò,Ò.J/J1Â1Â3:4:6²6²8*9*;¢;¢=>@’@’B C E‚E‚GúGúIpJpLæLæN\O\QÒQÒSHTHV¾V¾X4Y4[ª[ª] ^ `–`–b c e‚e‚gøgøinjnläländodqÚqÚsPtPvÆvÆx!?!I–I–S T ^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccØØMM 7 7 ¬ ¬!!––  €€õõjjßßT T"É"É$>%>'³'³)(*(,�,�./1‡1‡3ü3ü5q6q8æ8æ:[;[=Ð=Ð?E@EBºBºD/E/G¤G¤IJLŽLŽNOQxQxSíSíUbVbX×X×ZL[L]Á]Á_6`6b«b«d e g•g•i j llnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1~!1�~!1½¾!1 ½¾!1#Å!1#Å!1ñç!1€ñç!1�("1ü�("1õ."1õ."1ÃQ"1pÃQ"1a’"1ìa’"1ǘ"1ôǘ"1óØ"1`óØ"1Sý"1ÌSý"10#1ì0#1é~#1L é~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½k7 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ƒÃ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½k7 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ƒÃ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‰½k7 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ƒÃ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ãã  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÄÄ):*:4°4°>$?$I˜I˜S T ^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$L%L'Â'Â)8*8,®,®.$/$1š1š346†6†8ü8ü:r;r=è=è?^@^BÔBÔDJEJGÀGÀI4J4L¨L¨NOQ�Q�STVxVxXìXìZ`[`]Ô]Ô_H`Hb¼b¼d0e0g¤g¤ijlŒlŒnoq|q|sðsðudvdxØxØzL{L}À}À4€4‚¨‚¨„…‡�‡�‰ŠŒxŒxŽìŽì�`‘`“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1§}!1�§}!1E¾!1 E¾!1¡Â!1¡Â!1sç!1€sç!1("1ü("1s,"1s,"1EQ"1pEQ"1ã‘"1ìã‘"1E–"1ôE–"11Ú"1`1Ú"1‘þ"1Ô‘þ"10#1ô0#1é~#1T é~#1Ï$1´ Ï$1÷ð$1x ÷ð$1ÕÀ%1| ÕÀ%1�&1\ �&1ûP'1 ûP'1£c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠõÁ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠõÁ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠõÁ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1Õe¸$  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee ß ßYYÑÑ)G*G4½4½>1?1I¥I¥ST^™^™h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯))££ — —"#%‰%‰'ÿ'ÿ)u*u,ë,ë.a/a1×1×3M4M6Ã6Ã8999;¯;¯=%>%@›@›BCE‡E‡GýGýIqJqLåLåNYOYQÍQÍSATAVµVµX)Y)[�[�]^`…`…bùbùdmemgágáiUjUlÉlÉnIoIq½q½s1t1v¥v¥xy{�{�}~€u€u‚é‚é„]…]‡чщEŠEŒ¹Œ¹Ž-�-‘¡‘¡“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!1~!1�~!1½¾!1 ½¾!15Å!15Å!1ñç!1€ñç!1�("1ü�("1/"1/"1ÃQ"1pÃQ"1a’"1ìa’"1Ù˜"1ôÙ˜"11Ú"1`1Ú"1‘þ"1Ô‘þ"10#1ô0#1é~#1T é~#1Õ$1´ Õ$1ýð$1x ýð$1ÛÀ%1| ÛÀ%1 �&1\  �&1Q'1 Q'1©c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Ær> b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1–Ö€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Ær> b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1–Ö€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Ær> b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1–Ö €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1¾q%  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ Õ ÕLLÍÍ)D*D4»4»>0?0I¥I¥ST^›^›his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÕÕLLÃà : : ± ±((ŸŸ��{{òòi i"à"à$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubêbêd_e_gÔgÔiIjIl¾l¾n?o?q´q´s)t)vžvžxy{ˆ{ˆ}ý}ýr€r‚ç‚ç„\…\‡чщFŠFŒ»Œ»Ž0�0‘¥‘¥“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!1¡}!1�¡}!1?¾!1 ?¾!1¡Â!1¡Â!1sç!1€sç!1("1ü("1s,"1s,"1EQ"1pEQ"1ã‘"1ìã‘"1E–"1ôE–"1[Ú"1`[Ú"1»þ"1Ô»þ"10#1ô0#1é~#1T é~#1·$1´ ·$1ßð$1x ßð$1½À%1| ½À%1“€&1\ “€&1‰P'1 ‰P'11c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆñ½ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆñ½ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Tˆñ½ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1üŒßK  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSËË)?*?4³4³>%?%I—I—S T ^‡^‡hùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnïnïpaqasÓsÓuEvEx·x·z){)}›}› € ‚‚„ñ„ñ†c‡c‰Õ‰Õ‹GŒG޹޹�+‘+“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!1%~!1�%~!1þ!1 þ!15Å!15Å!1ñç!1€ñç!1�("1ü�("1/"1/"1ÃQ"1pÃQ"1a’"1ìa’"1Ù˜"1ôÙ˜"1‘Ú"1`‘Ú"1ñþ"1Ôñþ"10#1ô0#1é~#1T é~#1½$1´ ½$1åð$1x åð$1ÃÀ%1| ÃÀ%1Ÿ€&1\ Ÿ€&1•P'1 •P'1=c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Æt@ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1—×€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Æt@ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1—×€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1’Æt@ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1—× €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1è0σ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Õ ÕKKÈÈ)=*=4²4²>%?%I˜I˜S T ^Š^Šhýhýrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……ûûqqçç] ]"Ó"Ó$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnònòpeqesØsØuKvKx¾x¾z1{1}¤}¤€‚ЂЄý„ý†p‡p‰ã‰ã‹VŒVŽÉŽÉ�<‘<“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„+i!1_}!1_}!1ý½!1|ý½!1aÃ!1„aÃ!1—Æ!1ð—Æ!1!Ð!1|!Ð!1WÓ!1HWÓ!11ç!1Ô1ç!1Ï'"1PÏ'"13-"1X3-"1i0"1Äi0"1ó9"1Pó9"1)="1)="1Q"1¨Q"1¡‘"1$ ¡‘"1—"1, —"1;š"1˜ ;š"1Å£"1$ Å£"1û¦"1ð û¦"1=Ô"1| =Ô"1%Ú"1X%Ú"1…þ"1Ô…þ"1Ç+#1ôÇ+#1s:#1Ðs:#1/o#1\/o#1‘Ž#1€‘Ž#1�™#1¬�™#1Ï$18Ï$1§ò$1ü§ò$1áÞ%1áÞ%1[ì%1à[ì%1�&1l�&1ÕR'1°ÕR'1}e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P½‰ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Å bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P½‰ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Å  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P½‰  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Å bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ååi‰" bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1U• `ê*¿T  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dUÕ  *ªùyÿÒRÔT«+©)~þSÓ(¨ý}ÒR§' | ü Q Ñ & ¦ û { Ð P ¥ å ° ð :zÏd¤ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ Õ ÕÍÍ)D*D4»4»>2?2I§I§ST^�^�his‡s‡} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÕÕLLÃà : : ± ±((ŸŸ��ÌÌCCºº1 1"¨"¨$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbìbìdaeagÖgÖiKjKlÀlÀnAoAq¶q¶s+t+v v xy{Š{Š}ÿ}ÿt€t‚é‚é„^…^‡Ó‡Ó‰HŠHŒ½Œ½Ž2�2‘§‘§“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!17~!1è7~!1Õ¾!1dÕ¾!1‰Å!1l‰Å!1Ï!1èÏ!1IÒ!1´IÒ!1 è!1@ è!1§("1¼§("1[/"1Ä[/"1å8"1@å8"1<"1 <"1ÛQ"1˜ÛQ"1y’"1y’"1-™"1 -™"1·¢"1˜ ·¢"1í¥"1d í¥"1/Ó"1ð /Ó"1‹Ú"1Ì ‹Ú"1ëþ"1H ëþ"1-,#1h-,#1y:#1Dy:#15o#1Ð5o#1£Ž#1ô£Ž#1¡™#1 ¡™#1Ã$1¬Ã$1›ò$1p›ò$1ÕÞ%1tÕÞ%1Oì%1TOì%1É€&1àÉ€&1™R'1$™R'1Ae'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1®â�\ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•uµê* bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1®â�\ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•uµê* bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1®â�\  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•uµê*  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~ž,Lg‡ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1&fU•ê*�ÁX  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d&¦UÕ*ª…ÿädÔTÃC©)~þSÓ(¨ÿÖV­- „  [ Û 2 ²  ‰ à ` µ õ VJŠá!x¸ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛPP��)<*<4±4±>'?'I›I›ST^�^�hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››……úúo o"ä"ä$#%#'˜'˜) * ,‚,‚.÷.÷0¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`KbËbËd?e?g³g³i'j'l›l›noqƒqƒs÷s÷ukvkxßxßzS{S}Ç}Ç;€;‚¯‚¯„#…#‡™‡™‰ŠŒƒŒƒŽøŽø�m‘m“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„+i!1_}!1_}!1ý½!1|ý½!1aÃ!1„aÃ!1—Æ!1ð—Æ!1!Ð!1|!Ð!1WÓ!1HWÓ!1+ç!1Ô+ç!1É'"1PÉ'"13-"1X3-"1i0"1Äi0"1ó9"1Pó9"1)="1)="1ýP"1¨ýP"1›‘"1$ ›‘"1—"1, —"1;š"1˜ ;š"1Å£"1$ Å£"1û¦"1ð û¦"1=Ô"1| =Ô"1IÚ"1XIÚ"1©þ"1Ô©þ"1ë+#1ôë+#1s:#1Ðs:#1/o#1\/o#1‘Ž#1€‘Ž#1�™#1¬�™#1·$18·$1�ò$1ü�ò$1ÉÞ%1ÉÞ%1Cì%1àCì%1Ÿ€&1lŸ€&1oR'1°oR'1e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oº† b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Å bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oº† b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Å  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Oº†  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Å bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Å剩(H bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1U•ê*¹ù¿T  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dUÕ*ªÿÔT©)„~þ]ÝSÓ6¶(¨�ý}ÒR§' | ü Q Ñ & ¦ û { Ð P ¥ å :zI‰Ïd¤ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Õ ÕÈÈ)=*=4²4²>'?'IšIšS T ^Œ^Œhÿhÿrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……ÂÂ88®®$ $"š"š$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I+J+LžLžNOQ„Q„S÷S÷UjVjXÝXÝZP[P]Ã]Ã_6`6b©b©deg�g�ijlulunônôpgqgsÚsÚuMvMxÀxÀz3{3}¦}¦€‚Œ‚Œ„ÿ„ÿ†r‡r‰å‰å‹XŒXŽËŽË�>‘>“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx+i!1=~!1è=~!1Û¾!1dÛ¾!1‰Å!1l‰Å!1Ï!1èÏ!1IÒ!1´IÒ!1è!1@è!1­("1¼­("1[/"1Ä[/"1å8"1@å8"1<"1 <"1áQ"1˜áQ"1’"1’"1-™"1 -™"1·¢"1˜ ·¢"1í¥"1d í¥"1/Ó"1ð /Ó"1[Ú"1Ì [Ú"1»þ"1H »þ"1ý+#1hý+#1Ù:#1DÙ:#1•o#1Еo#1•�#1ô•�#1“›#1 “›#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1‡€&1à‡€&1WR'1$WR'1ÿd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1°ä”` b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•v¶ê* bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1°ä”` b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•v¶ê* bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1°ä”`  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•v¶ê*  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 –¶6V<\ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1p°Å `Î|¼  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dpð`àPÐ@À¢"0°�  `à�?¿€înÜ\ Ê J ¸ 8 ¦ & ”  ‚ pðµõ` P¾þl¬ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×LLŒŒ):*:4¯4¯>$?$I—I—ST^‡^‡hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡üüqqææ[ ["Ð"Ð$%'…'…)ú)ú+o,o.ä.ä0’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I$J$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b­b­deg‘g‘ijlulunçnçpYqYsËsËu=v=x¯x¯z!{!}“}“€‚w‚w„é„é†]‡]‰ЉЋCŒC޶޶�)‘)“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1¡}!1è¡}!1?¾!1d?¾!1‰Å!1l‰Å!1Ï!1èÏ!1IÒ!1´IÒ!1sç!1@sç!1("1¼("1[/"1Ä[/"1å8"1@å8"1<"1 <"1EQ"1˜EQ"1ã‘"1ã‘"1-™"1 -™"1·¢"1˜ ·¢"1í¥"1d í¥"1/Ó"1ð /Ó"1sÚ"1Ì sÚ"1Óþ"1H Óþ"1,#1h,#1Í:#1DÍ:#1‰o#1Љo#1;�#1ô;�#19›#1 9›#1Ã$1¬Ã$1›ò$1p›ò$1ÕÞ%1tÕÞ%1Oì%1TOì%1Û€&1àÛ€&1«R'1$«R'1Se'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 DU•ê* bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 DU•ê* bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ö  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 DU•ê*  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 nŽKk , bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1#ci©R½ýh¨  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d#£iéþ~TÔÙY=½´4&¦�øxáaÊJµ5  ‹ v ö a á L Ì 7 · Õ "¢ K´ô_Ÿ J bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ü ÜRRÈÈ)>*>4´4´>(?(IdIdST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÜÜVVÐÐ J J Ä Ä>>¸¸22¬¬""˜˜ „ „"ú"ú$p%p'æ'æ)\*\,Ò,Ò.H/H1¾1¾34446ª6ª8 9 ;–;–= > @€@€BôBôDhEhGÜGÜIPJPLÄLÄN8O8Q¬Q¬S T V\V\XÐXÐZD[D]¸]¸_,`,bâbâdVeVgÊgÊi>j>l²l²n&o&qšqšstv‚v‚xöxözj{j}Þ}ÞR€R‚ƂƄ:…:‡®‡®‰!Š!Œ”Œ”Ž�‘z‘z“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!11~!1è1~!1Ͼ!1dϾ!1‰Å!1l‰Å!1Ï!1èÏ!1IÒ!1´IÒ!1è!1@è!1¡("1¼¡("1[/"1Ä[/"1å8"1@å8"1<"1 <"1ÕQ"1˜ÕQ"1s’"1s’"1-™"1 -™"1·¢"1˜ ·¢"1í¥"1d í¥"1/Ó"1ð /Ó"1¯Ú"1Ì ¯Ú"1ÿ"1H ÿ"1Q,#1hQ,#1y:#1Dy:#15o#1Ð5o#1‘Ž#1ô‘Ž#1�™#1 �™#1Ã$1¬Ã$1›ò$1p›ò$1ÕÞ%1tÕÞ%1Oì%1TOì%1í€&1àí€&1½R'1$½R'1ee'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1£×…Q b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•§çê* bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1£×…Q b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•§çê* bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1£×…Q  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•§çê*  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �¡2R„¤ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1YU•ê*¿T  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d™UÕ*ªövÿÓSÔT°0©)~þSÓ(¨ý}ÒR§' | ü Q Ñ & ¦ û { Ð P ¥ å é ):zÏd¤ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>/?/I¦I¦ST^Y^Yhisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIWJWLÎLÎNEOEQ¼Q¼S3T3VªVªX!Y![˜[˜]^`†`†býbýdtetgëgëibjblÙlÙnoq‡q‡sùsùukvkxÝxÝz‡{‡}ù}ùk€k‚݂݄O…O‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1¡}!1è¡}!1?¾!1d?¾!1‰Å!1l‰Å!1Ï!1èÏ!1IÒ!1´IÒ!1sç!1@sç!1("1¼("1[/"1Ä[/"1å8"1@å8"1<"1 <"1EQ"1˜EQ"1ã‘"1ã‘"1-™"1 -™"1·¢"1˜ ·¢"1í¥"1d í¥"1/Ó"1ð /Ó"1UÚ"1Ì UÚ"1µþ"1H µþ"1÷+#1h÷+#1©:#1D©:#1eo#1Ðeo#1��#1ô��#1š#1 š#1«$1¬«$1ƒò$1pƒò$1½Þ%1t½Þ%17ì%1T7ì%1c€&1àc€&13R'1$3R'1Ûd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ôÀ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 CU•ê* bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ôÀ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 CU•ê* bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰ôÀ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 CU•ê*  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �°>0P bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1b¢¾þD¤äD„  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dbâDÄ&¦ˆ�êjjêÌLEÅ®.  �pðPÐ0 °  � ð p Ð P ° 0 � r²p°T´ôT” bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚNNÂÂ)6*64ª4ª>?IUIUST^z^zhìhìr^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""šš‚‚ööj j"Þ"Þ$R%R'Æ'Æ):*:,®,®."/"1–1–3 4 6~6~8ò8ò:f;f=Ú=Ú?L@LB¾B¾D0E0G¢G¢IJL†L†NøNøPjQjSÜSÜUVX‡X‡ZùZù\k]k_Ý_Ýa�b�degtgtiæiækXlXnÊnÊp‡>‰¯‰¯‹ Œ ޑޑ�‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!17~!1è7~!1Õ¾!1dÕ¾!1‰Å!1l‰Å!1Ï!1èÏ!1IÒ!1´IÒ!1è!1@è!1¡("1¼¡("1[/"1Ä[/"1å8"1@å8"1<"1 <"1ÕQ"1˜ÕQ"1s’"1s’"1-™"1 -™"1·¢"1˜ ·¢"1í¥"1d í¥"1/Ó"1ð /Ó"1…Ú"1Ì …Ú"1åþ"1H åþ"1',#1h',#1y:#1Dy:#15o#1Ð5o#1‘Ž#1ô‘Ž#1�™#1 �™#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1“€&1à“€&1cR'1$cR'1 e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¥Ù‰U b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•¦æê* bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¥Ù‰U b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•¦æê* bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¥Ù‰U  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•¦æê*  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 •µ[{:Z bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1U•ê*¶ö¿T  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dUÕ*ªÿÔT©)� ~þjêSÓGÇ(¨$¤ý}ÒR§' | ü Q Ñ & ¦ û { Ð P ¥ å :z†ÆÏd¤ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ö ÖKKÀÀ)5*54ª4ª>!?!I–I–S T ^H^Hhóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÖÖKKÀÀ 5 5 ª ª””  ~~óóhhÝÝR R"Ç"Ç$<%<'±'±)&*&,›,›./1…1…3ú3ú5o6o8ä8ä:Y;Y=Î=Î?C@CB¸B¸D-E-G¢G¢IJLŽLŽNOQxQxSíSíUbVbX×X×ZL[L]Á]Á_6`6b«b«d e g•g•i j lln¼n¼p.q.s s uvx„x„z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1§}!1Ч}!1E¾!1LE¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1yç!1(yç!1("1¤("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1KQ"1€KQ"1é‘"1üé‘"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1{2#1,{2#1�#1˜�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1)�&1Ì)�&1ùR'1ùR'1¡e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ÷à b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 TU•ê* bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ÷à b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 TU•ê* bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠ÷à  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 TU•ê*  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 s“6hˆ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1\€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1œùyÖV³3ì , €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Û ÛTTÍÍ)F*F4¿4¿>8?8I¯I¯S$T$^—^—h i s}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÛÛTTÍÍ F F ¿ ¿88±±**££•• ‡ ‡"#%y%y'ò'ò)k*k,ä,ä.]/]1Ö1Ö3O4O6È6È8A9A;º;º=3>3@¬@¬B%C%EžEžGHJ�J�LMO~O~QõQõSlTlVãVãXZYZ[Ñ[Ñ]H^H`¿`¿b4c4e©e©ghj“j“lmo{o{qîqîsatavÔvÔxGyG{º{º}-~-€ € ‚ƒ…†…†‡ù‡ù‰kŠkŒ݌ݎO�O‘Á‘Á“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1+~!1Ð+~!1ɾ!1Lɾ!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ýç!1(ýç!1›("1¤›("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1ÏQ"1€ÏQ"1m’"1üm’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1ÁÚ"1´ ÁÚ"1!ÿ"10 !ÿ"1c,#1Pc,#1o2#1,o2#1çŒ#1˜çŒ#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ÿ€&1Ìÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜ÌzF b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•�Áî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜ÌzF b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•�Áî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜ÌzF  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•�Áî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 „¤8X•µ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 L€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ŒçgÂB�¼ ü  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á á[[ÑÑ)G*G4½4½>3?3I§I§ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±++¥¥ ™ ™"#%‰%‰'ÿ'ÿ)u*u,ë,ë.a/a1×1×3M4M6Ã6Ã8999;¯;¯=%>%@›@›BCE‡E‡GýGýIsJsLçLçN[O[QÏQÏSCTCV·V·X+Y+[Ÿ[Ÿ]^`‡`‡bûbûdoeogãgãiWjWlËlËn?o?q³q³s't'v›v›xy{ƒ{ƒ}÷}÷k€k‚߂߄S…S‡LJlj;Š;Œ¯Œ¯Ž#�#‘—‘—“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1§}!1Ч}!1E¾!1LE¾!1�Å!1T�Å!1Ï!1ÐÏ!1OÒ!1œOÒ!1sç!1(sç!1("1¤("1a/"1¬a/"1ë8"1(ë8"1!<"1ô!<"1EQ"1€EQ"1ã‘"1üã‘"13™"1 3™"1½¢"1€ ½¢"1ó¥"1L ó¥"15Ó"1Ø 5Ó"1sÚ"1´ sÚ"1Óþ"10 Óþ"1,#1P,#15#1,5#1¹�#1˜¹�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1·€&1Ì·€&1‡R'1‡R'1/e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠõÁ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UU•ê* bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠõÁ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UU•ê* bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1VŠõÁ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UU•ê*  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‘±Cc2R bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¹ù€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�míJÊ'§‰É €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÇÇ)>*>4µ4µ>,?,I¡I¡ST^…^…höhörhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbäbädWeWgÊgÊi=j=l°l°n!o!q’q’stvtvtxåxåzV{V}Ç}Ç8€8‚©‚©„…‡Œ‡Œ‰ý‰ý‹nŒnŽߎß�P‘P“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!11~!1Ð1~!1Ͼ!1LϾ!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ýç!1(ýç!1›("1¤›("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1ÏQ"1€ÏQ"1m’"1üm’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1©Ú"1´ ©Ú"1 ÿ"10  ÿ"1K,#1PK,#1 5#1, 5#1ƒ�#1˜ƒ�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎ~J b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•�Áï/ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎ~J b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•�Áï/ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎ~J  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•�Áï/  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ”´8Xzš bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1§ç€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1xøSÓ.® ‰W— €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝUUÉÉ)=*=4±4±>%?%I—I—S T ^{^{híhír_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%���� } }"õ"õ$i%i'Ý'Ý)Q*Q,Å,Å.9/91­1­3!4!6•6•8 9 ;};}=ñ=ñ?e@eBÙBÙDMEMGÁGÁI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnãnãpUqUsÇsÇu9v9x«x«z{}�}�€‚s‚s„å„å†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1­}!1Э}!1K¾!1LK¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ç!1(ç!1("1¤("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1QQ"1€QQ"1ï‘"1üï‘"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1©Ú"1´ ©Ú"1 ÿ"10  ÿ"1K,#1PK,#1{2#1,{2#15�#1˜5�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1;�&1Ì;�&1 S'1 S'1³e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹øÄ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @W—ì, bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹øÄ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @W—ì, bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹øÄ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @W—ì,  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x˜ @y™ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1S€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1“òrÑQ°0C €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚRRÊÊ)B*B4º4º>2?2I¨I¨ST^”^”h i s€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""ššŠŠ z z"ò"ò$j%j'â'â)Z*Z,Ò,Ò.J/J1Â1Â3:4:6²6²8*9*;¢;¢=>@’@’B C E‚E‚GúGúIrJrLèLèN^O^QÔQÔSJTJVÀVÀX6Y6[¬[¬]"^"`˜`˜bce„e„gúgúipjplælæn\o\qÒqÒsHtHv¾v¾x4y4{ª{ª} ~ €–€–‚ ƒ …‚…‚‡ø‡ø‰nŠnŒäŒäŽZ�Z‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1%~!1Ð%~!1þ!1Lþ!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1ñç!1(ñç!1�("1¤�("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1ÃQ"1€ÃQ"1a’"1üa’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1mÚ"1´ mÚ"1Íþ"10 Íþ"1,#1P,#1c2#1,c2#1ÉŒ#1˜ÉŒ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Áo; b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•œÜù9 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Áo; b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•œÜù9 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Áo;  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•œÜù9  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‡§Åå>^ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÚZ´4Žš Ú  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94®4®>!?!I”I”ST^z^zhíhír`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?M@MBÀBÀD3E3G¦G¦IJLŒLŒNÿNÿPrQrSåSåUXVXXËXËZ>[>]±]±_$`$b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†T‡T‰ƉƋ8Œ8ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1­}!1Э}!1K¾!1LK¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1yç!1(yç!1("1¤("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1KQ"1€KQ"1é‘"1üé‘"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#15#1,5#1Ñ�#1˜Ñ�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @V–ë+ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @V–ë+ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹ö  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @V–ë+  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ’²b‚4T bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1²ò€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�nîMÍ,¬¢â €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÄÄ):*:4°4°>&?&IšIšST^‚^‚höhörjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$L%L'Â'Â)8*8,®,®.$/$1š1š346†6†8ü8ü:r;r=è=è?^@^BÔBÔDJEJGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqvqvsêsêu^v^xÒxÒzF{F}º}º.€.‚¢‚¢„…‡ЇЉþ‰þ‹rŒr޿޿�Z‘Z“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1%~!1Ð%~!1þ!1Lþ!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1÷ç!1(÷ç!1•("1¤•("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1ÉQ"1€ÉQ"1g’"1üg’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1aÚ"1´ aÚ"1Áþ"10 Áþ"1,#1P,#1ý4#1,ý4#1e�#1˜e�#1Ÿ$1$Ÿ$1wò$1èwò$1±Þ%1ì±Þ%1!€&1Ì!€&1ñQ'1ñQ'1™d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ãs? b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•›Ûú: bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ãs? b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•›Ûú: bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ãs?  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•›Ûú:  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 “³Åå6V bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1šÚ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1jêDÄžøx4t €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚMMÀÀ)3*34¦4¦>?IˆIˆSùSù]j^jhÛhÛrLsL} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠýýppããV V"É"É$<%<'¯'¯)"*",•,•./1{1{3î3î5a6a8Ô8Ô:G;G=º=º?+@+BœBœD E G~G~IïIïK`L`NÑNÑPBQBS³S³U$V$X•X•Z[]w]w_è_èaYbYdÊdÊf;g;i¬i¬klnŽnŽpÿpÿrpspuáuáwRxRzÃzÃ|4}4¥¥�‚„‡„‡†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1³}!1г}!1Q¾!1LQ¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1ç!1(ç!1("1¤("1/"1¬/"1 9"1( 9"1?<"1ô?<"1QQ"1€QQ"1ï‘"1üï‘"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1©Ú"1´ ©Ú"1 ÿ"10  ÿ"1K,#1PK,#1u2#1,u2#1S�#1˜S�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1�&1Ì�&1ÕR'1ÕR'1}e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒùÅ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Im­B bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒùÅ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Im­B bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒùÅ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Im­B  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 }�*Jr’ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 K€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ‹ìlÍM®.[ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÇÇ)>*>4µ4µ>,?,I¡I¡ST^‹^‹hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbæbæd[e[gÐgÐiEjElºlºn/o/q¤q¤stvŽvŽxy{x{x}í}íb€b‚ׂׄL…L‡Á‡Á‰6Š6Œ«Œ«Ž � ‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1~!1Ð~!1½¾!1L½¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1ëç!1(ëç!1‰("1¤‰("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1½Q"1€½Q"1[’"1ü[’"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1i2#1,i2#1w�#1˜w�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶d0 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•�Áí- bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶d0 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•�Áí- bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶d0  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•�Áí-  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‚¢Åå4T bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€åeÊJ¯/@€ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßTTÉÉ)>*>4³4³>)?)I�I�ST^…^…hùhùroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ‰‰þþs s"è"è$]%]'Ò'Ò)G*G,¼,¼.1/11¦1¦346�6�89;z;z=ï=ï?d@dBÙBÙDNENGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒyŒyŽîŽî�c‘c“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1­}!1Э}!1K¾!1LK¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1ç!1(ç!1("1¤("1/"1¬/"1 9"1( 9"1?<"1ô?<"1QQ"1€QQ"1ï‘"1üï‘"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#15#1,5#1ï�#1˜ï�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1Ÿ€&1ÌŸ€&1oR'1oR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ÷à b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jl¬A bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ÷à b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jl¬A bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒ÷à  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jl¬A  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 “³cƒ6V bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¬ì€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�pðQÑ2²¼ü €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙOOÄÄ)9*94®4®>#?#I–I–S T ^|^|hïhïrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰ÿÿuuëëa a"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnänäpWqWsÊsÊu=v=x°x°z#{#}–}– € ‚|‚|„ï„ï†b‡b‰Õ‰Õ‹HŒH޻޻�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1~!1Ð~!1½¾!1L½¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1ñç!1(ñç!1�("1¤�("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1ÃQ"1€ÃQ"1a’"1üa’"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1[Ú"1´ [Ú"1»þ"10 »þ"1ý+#1Pý+#15#1,5#1Ž#1˜Ž#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸h4 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•�Áî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸h4 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•�Áî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸h4  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•�Áî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ž®Åå,L bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¥å€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1”yù^ÞCÃå% €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛPPÅÅ):*:4¯4¯>$?$I—I—S T ^{^{híhírasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹uuêê_ _"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I$J$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b¡b¡deg…g…i÷i÷kilinÛnÛpMqMs¿s¿u1v1x£x£z{}‡}‡ùù�k‚k„݄݆Q‡Q‰ĉċ7Œ7ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1¹}!1й}!1W¾!1LW¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1…ç!1(…ç!1#("1¤#("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1WQ"1€WQ"1õ‘"1üõ‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#1o2#1,o2#1M�#1˜M�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ç€&1Ìç€&1·R'1·R'1_e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�úÆ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @V–ë+ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�úÆ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @V–ë+ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�úÆ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @V–ë+  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Å倠0P bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1„åeÆF§'T €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff à àVVÌÌ)B*B4¸4¸>,?,I I ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffààZZÔÔ N N È ÈBB¼¼66°°&&œœ ˆ ˆ"þ"þ$t%t'ê'ê)`*`,Ö,Ö.L/L1Â1Â38486®6®8$9$;š;š=>@„@„BøBøDlElGàGàITJTLÈLÈN3?3IªIªS!T!^“^“hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâI[J[LÒLÒNIOIQÀQÀS7T7V®V®X%Y%[œ[œ]^`Š`ŠbcexexgïgïifjflÝlÝnOoOqÁqÁs3t3v¥v¥xy{‰{‰}û}ûm€m‚߂߄Q…Q‡ÇÉ5Š5Œ§Œ§Ž�‘‹‘‹“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1³}!1г}!1Q¾!1LQ¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1…ç!1(…ç!1#("1¤#("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1WQ"1€WQ"1õ‘"1üõ‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1[Ú"1´ [Ú"1»þ"10 »þ"1ý+#1Pý+#15#1,5#1ï�#1˜ï�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1o€&1Ìo€&1?R'1?R'1çd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�øÄ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @V–ë+ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�øÄ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @V–ë+ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�øÄ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @V–ë+  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Å努$D bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¥å€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ˆiéJÊ+«µõ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞRRÆÆ):*:4®4®> ? I’I’ST^v^vhèhèrZsZ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žž††úún n"â"â$V%V'Ê'Ê)>*>,²,².&/&1š1š346‚6‚8ö8ö:j;j=Þ=Þ?P@PBÂBÂD4E4G¦G¦IJLŠLŠNüNüPnQnSàSàURVRXÄXÄZ6[6]¨]¨_`bŒbŒdþdþfpgpiâiâkTlTnÆnÆp8q8sªsªuvxŽxŽz{}r}rää�V‚V„ȄȆ:‡:‰«‰«‹ŒŽ�Ž��þ�þ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1~!1Ð~!1·¾!1L·¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1ëç!1(ëç!1‰("1¤‰("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1½Q"1€½Q"1[’"1ü[’"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#15#1,5#1ÙŽ#1˜ÙŽ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1¥€&1Ì¥€&1uR'1uR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­]) b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U••Õë+ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­]) b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U••Õë+ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­])  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U••Õë+  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Åå�¡2 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1°ð€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1À@°0  � à €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ú ÚOOÄÄ)9*94®4®>%?%IšIšST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÚÚOOÄÄ 9 9 ® ®##˜˜  ‚‚÷÷llááV V"Ë"Ë$@%@'µ'µ)***,Ÿ,Ÿ./1‰1‰3þ3þ5s6s8è8è:];]=Ò=Ò?G@GB¼B¼D1E1G¦G¦IJL’L’NOQ|Q|SñSñUfVfXÛXÛZP[P]Å]Å_:`:b¯b¯d$e$g™g™ijlƒlƒnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1¹}!1й}!1W¾!1LW¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1‹ç!1(‹ç!1)("1¤)("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1]Q"1€]Q"1û‘"1üû‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1[Ú"1´ [Ú"1»þ"10 »þ"1ý+#1Pý+#1o2#1,o2#1Ý�#1˜Ý�#1á$1$á$1¹ò$1è¹ò$1óÞ%1ìóÞ%1;�&1Ì;�&1 S'1 S'1³e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽûÇ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @V–ë+ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽûÇ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @V–ë+ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽûÇ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @V–ë+  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ååy™"B bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€éiÒR»; à €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff ß ßXXÑÑ)J*J4Ã4Ã>7@°@°B)C)E¢E¢GHJ”J”L M O‚O‚QùQùSpTpVçVçX^Y^[Õ[Õ]L^L`Ã`Ãb8c8e­e­g"h"j—j—l m ooqòqòsetevØvØxKyK{¾{¾}1~1€¤€¤‚ƒ…ЅЇý‡ý‰oŠoŒáŒáŽS�S‘őœ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1 ~!1Ð ~!1«¾!1L«¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1ßç!1(ßç!1}("1¤}("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1±Q"1€±Q"1O’"1üO’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#1o2#1,o2#1ߎ#1˜ߎ#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1�&1Ì�&1áR'1áR'1‰e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l N b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•ƒÃê* bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l N b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•ƒÃê* bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l N  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•ƒÃê*  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Åår’2 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€û{övñq à €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk å å__ÕÕ)K*K4Á4Á>7?7I«I«ST^“^“his{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkåå__ÙÙ S S Í ÍGGÁÁ;;µµ//©©## � �"#%�%�'(*y*y,ï,ï.e/e1Û1Û3Q4Q6Ç6Ç8=9=;³;³=)>)@Ÿ@ŸBCE‹E‹GHJwJwLëLëN_O_QÓQÓSGTGV»V»X/Y/[£[£]^`‹`‹bÿbÿdsesgçgçi[j[lÏlÏnCoCq·q·s+t+vŸvŸxy{‡{‡}û}ûo€o‚ã‚ã„W…W‡ˇˉ?Š?Œ³Œ³Ž'�'‘›‘›“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1¹}!1й}!1W¾!1LW¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1‹ç!1(‹ç!1)("1¤)("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1]Q"1€]Q"1û‘"1üû‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1IÚ"1´ IÚ"1©þ"10 ©þ"1ë+#1Pë+#15#1,5#1…Ž#1˜…Ž#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽùÅ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @V–ë+ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽùÅ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @V–ë+ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽùÅ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @V–ë+  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÅåŸ. bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1©é€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¤$� vö_ßI‰ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ý ÝTTËË)B*B4¹4¹>0?0I¥I¥ST^‰^‰húhúrlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÝÝTTËË B B ¹ ¹00§§••  ƒƒúúq q"è"è$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n%o%q–q–stvxvxxéxézZ{Z}Ë}Ë<€<‚­‚­„…‡�‡�‰ŠŒrŒrŽãŽã�T‘T“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1~!1Ð~!1±¾!1L±¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1åç!1(åç!1ƒ("1¤ƒ("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1·Q"1€·Q"1U’"1üU’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"11Ú"1´ 1Ú"1‘þ"10 ‘þ"1Ó+#1PÓ+#1-5#1,-5#1™�#1˜™�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢R b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•‚Âê* bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢R b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U•‚Âê* bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢R  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U•‚Âê*  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ååt”ø bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1»û€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ìlçgâbÝ][› €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áYYÍÍ)A*A4µ4µ>)?)I›I›S T ^^hñhñrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‘‘  � �"ù"ù$m%m'á'á)U*U,É,É.=/=1±1±3%4%6™6™8 9 ;�;�=õ=õ?i@iBÝBÝDQEQGÅGÅI9J9L«L«NOQ�Q�STVsVsXåXåZW[W]É]É_;`;b­b­deg‘g‘ijlulunçnçpYqYsËsËu=v=x¯x¯z!{!}“}“€‚w‚w„é„é†[‡[‰͉͋?Œ?ޱޱ�#‘#“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1¿}!1п}!1]¾!1L]¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1‘ç!1(‘ç!1/("1¤/("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1cQ"1€cQ"1’"1ü’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#1u2#1,u2#1‘Ž#1˜‘Ž#1ç$1$ç$1¿ò$1è¿ò$1ùÞ%1ìùÞ%1M�&1ÌM�&1S'1S'1Åe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�üÈ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @V–ë+ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�üÈ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @V–ë+ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�üÈ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @V–ë+  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Åår’4 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ôtèhÜ\P� €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÎÎ)F*F4¾4¾>6?6I¬I¬S"T"^˜^˜his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$n%n'æ'æ)^*^,Ö,Ö.N/N1Æ1Æ3>4>6¶6¶8.9.;¦;¦=>@–@–BCE†E†GþGþIvJvLìLìNbObQØQØSNTNVÄVÄX:Y:[°[°]&^&`œ`œbceˆeˆgþgþitjtlêlên`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1~!1Ð~!1¥¾!1L¥¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1Ùç!1(Ùç!1w("1¤w("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1«Q"1€«Q"1I’"1üI’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1u2#1,u2#1“�#1˜“�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•C b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U••Õë+ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•C b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U••Õë+ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•C  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U••Õë+  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 g‡þˆ¨ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€† Œ’P� €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞSSÈÈ)=*=4²4²>%?%I˜I˜S T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽxxííb b"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?Q@QBÄBÄD7E7GªGªIJL�L�NOQvQvSéSéU\V\XÏXÏZB[B]µ]µ_(`(b›b›deg�g�iôiôkglgnÚnÚpMqMsÀsÀu3v3x¦x¦z{}Œ}Œÿÿ�r‚r„å„å†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1¿}!1п}!1]¾!1L]¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1‘ç!1(‘ç!1/("1¤/("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1cQ"1€cQ"1’"1ü’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"17Ú"1´ 7Ú"1—þ"10 —þ"1Ù+#1PÙ+#1'5#1,'5#1E�#1˜E�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1Û€&1ÌÛ€&1«R'1«R'1Se'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�úÆ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @V–ë+ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�úÆ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @V–ë+ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�úÆ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @V–ë+  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ååt”ø bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1´ô€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÐPÄD¸8¬,D €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜRRÈÈ)>*>4´4´>*?*IžIžST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒxxîîd d"Ú"Ú$P%P'Æ'Æ)<*<,²,².(/(1ž1ž346Š6Š89;v;v=ì=ì?b@bBØBØDNENGÄGÄI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1 ~!1Ð ~!1«¾!1L«¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1Ùç!1(Ùç!1w("1¤w("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1«Q"1€«Q"1I’"1üI’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1ÁÚ"1´ ÁÚ"1!ÿ"10 !ÿ"1c,#1Pc,#195#1,95#1Y�#1˜Y�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%19€&1Ì9€&1 R'1 R'1±d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—G b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U••Õë+ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—G b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U••Õë+ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—G  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U••Õë+  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 g‡Þþ�¡ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Æ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1˜ž$¤*ªV €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞQQÄÄ)7*74ª4ª>?IŒIŒSýSý]n^nhßhßrPsP} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽttççZ Z"Í"Í$@%@'³'³)&*&,™,™. / 113ò3ò5e6e8Ø8Ø:K;K=¾=¾?/@/B B DEG‚G‚IóIóKdLdNÕNÕPFQFS·S·U(V(X™X™Z [ ]{]{_ì_ìa]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ü†üˆl‰l‹Ü‹Ü�LŽL�¼�¼’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Ý}!1�Ý}!1{¾!1{¾!1¹Â!1$¹Â!1©ç!1�©ç!1G("1G("1‹,"1$‹,"1{Q"1�{Q"1’"1’"1]–"1$]–"1Ù"1�Ù"1qý"1üqý"150#1 50#1#1| #1ó$1Ü ó$1Ëò$1  Ëò$1ß%1¤ ß%1)�&1„ )�&1ùR'1ÈùR'1¡e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¡ÚÓŸ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¡ÚÓŸ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations19m¡ÚÓŸ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Åå  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww î îeeÜÜ)S*S4Ê4Ê>???I´I´S)T)^ž^žhisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwîîeeÜÜ S S Ê ÊAA¸¸//¦¦””  ‚ ‚"ù"ù$p%p'ç'ç)^*^,Õ,Õ.L/L1Ã1Ã3:4:6±6±8(9(;Ÿ;Ÿ=>@�@�BCE{E{GòGòIgJgLÜLÜNQOQQÆQÆS;T;V°V°X%Y%[š[š]^`„`„bùbùdnengãgãiXjXlÍlÍnBoBq·q·s,t,v¡v¡xy{‹{‹}~€u€u‚ê‚ê„_…_‡Ô‡Ô‰IŠIŒ¾Œ¾Ž3�3‘¨‘¨“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1Ÿ¾!1 Ÿ¾!1SÅ!1SÅ!1Óç!1€Óç!1q("1üq("1%/"1%/"1¥Q"1p¥Q"1C’"1ìC’"1÷˜"1ô÷˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"150#1ì50#1#1L #1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1í€&1T í€&1½R'1˜½R'1ee'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰7 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1™Ù€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰7 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1™Ù€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰7 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1™Ù €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1zš  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ì ìaaÖÖ)K*K4À4À>4?4I¨I¨ST^�^�hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttììddÜÜ T T Ì ÌDD¼¼44¬¬!!––  € €"õ"õ$j%j'ß'ß)T*T,É,É.>/>1³1³3(4(6�6�89;‡;‡=ü=ü?q@qBæBæD[E[GÐGÐIDJDL¸L¸N,O,Q Q STVˆVˆXüXüZp[p]ä]ä_X`XbÌbÌd@e@g´g´i(j(lœlœnoq„q„søsøulvlxàxàzT{T}È}È<€<‚°‚°„$…$‡˜‡˜‰ Š Œ‚Œ‚Ž÷Ž÷�l‘l“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1×}!1�×}!1u¾!1u¾!1¹Â!1$¹Â!1©ç!1�©ç!1G("1G("1‹,"1$‹,"1{Q"1�{Q"1’"1’"1]–"1$]–"1Ù"1�Ù"1qý"1üqý"150#1 50#1#1| #1á$1Ü á$1¹ò$1  ¹ò$1óÞ%1¤ óÞ%1À&1„ À&1“R'1È“R'1;e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:kŸ ÖÖ¢ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:kŸ ÖÖ¢ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1:kŸ ÖÖ¢ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Åå  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ì ìbb××)L*L4Á4Á>4?4I§I§ST^�^�hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvììbbØØ N N Ä Ä::°°&&œœˆˆþþt t"ê"ê$_%_'Ô'Ô)I*I,¾,¾.3/31¨1¨346’6’89;|;|=ñ=ñ?f@fBÛBÛDPEPGÅGÅI8J8L«L«NOQ‘Q‘STVwVwXêXêZ][]]Ð]Ð_C`Cb¶b¶d)e)gœgœijl‚l‚nõnõphqhsÛsÛuNvNxÁxÁz4{4}§}§€‚�‚�„…‡s‡s‰æ‰æ‹YŒYŽÌŽÌ�?‘?“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1¥¾!1 ¥¾!1SÅ!1SÅ!1Óç!1€Óç!1q("1üq("1%/"1%/"1¥Q"1p¥Q"1C’"1ìC’"1÷˜"1ô÷˜"1ÍÚ"1`ÍÚ"1-ÿ"1Ì-ÿ"150#1ì50#1#1L #1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰9 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1˜Ø€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰9 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1˜Ø€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1U‰9 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1˜Ø €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1…¥  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ê ê__ÔÔ)I*I4¾4¾>1?1I¤I¤ST^ˆ^ˆhúhúrlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttêê``ÖÖ L L  Â88®®$$šš„„ùùn n"ã"ã$X%X'Í'Í)B*B,·,·.,/,1¡1¡346‹6‹89;u;u=ê=ê?_@_BÔBÔDIEIG¾G¾I1J1L¤L¤NOQŠQŠSýSýUpVpXãXãZV[V]É]É_<`v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰ωϋBŒB޵޵�(‘(“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Ñ}!1�Ñ}!1o¾!1 o¾!1¹Â!1¹Â!1£ç!1€£ç!1A("1üA("1‹,"1‹,"1uQ"1puQ"1’"1ì’"1]–"1ô]–"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"150#1ì50#1#1L #1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jž × b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jž × b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jž × b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1d„  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww ñ ñggÝÝ)S*S4Ç4Ç>;?;I¯I¯S#T#^—^—h i s}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwññkkåå _ _ Ù ÙSSÍÍGGÁÁ77­­## ™ ™"#%…%…'û'û)q*q,ç,ç.]/]1Ó1Ó3I4I6¿6¿8595;«;«=>@“@“BCE{E{GïGïIcJcL×L×NKOKQ¿Q¿S3T3V§V§XY[�[�]^`w`wbëbëd_e_gÓgÓiGjGl»l»n/o/q£q£stv‹v‹xÿxÿzs{s}ç}ç[€[‚ςτC…C‡µ‡µ‰(Š(Œ›Œ›Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1¥¾!1 ¥¾!1MÅ!1MÅ!1Ùç!1€Ùç!1w("1üw("1/"1/"1«Q"1p«Q"1I’"1ìI’"1ñ˜"1ôñ˜"1»Ú"1`»Ú"1ÿ"1Ìÿ"150#1ì50#1#1L #1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’@  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1•Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’@  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1•Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’@  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1•Õ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1u•  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp ç ç^^ÕÕ)L*L4Ã4Ã>:?:I±I±S(T(^š^šh i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppçç^^ÕÕ L L à Ã::±±((ŸŸ�� { {"ò"ò$i%i'à'à)W*W,Î,Î.E/E1¼1¼33436ª6ª8!9!;˜;˜=>@†@†BýBýDtEtGëGëIbJbLÙLÙNPOPQÇQÇS>T>VµVµX,Y,[£[£]^`‘`‘bceegögöimjmlälänVoVqÈqÈs:t:v¬v¬xy{�{�}~€t€t‚æ‚æ„X…X‡ʇʉ<Š<Œ®Œ®Ž � ‘’‘’“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Ñ}!1�Ñ}!1o¾!1 o¾!1¹Â!1¹Â!1�ç!1€�ç!1;("1ü;("1‹,"1‹,"1oQ"1poQ"1 ’"1ì ’"1]–"1ô]–"1µÚ"1`µÚ"1ÿ"1Ìÿ"150#1ì50#1#1L #1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœÓ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœÓ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœÓ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1o�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu í íaaÕÕ)I*I4»4»>-?-IŸIŸST^ƒ^ƒhõhõrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuííeeÝÝ U U Í ÍEE½½55­­!!••  } }"ñ"ñ$e%e'Ù'Ù)M*M,Á,Á.5/51©1©346‘6‘89;y;y=ë=ë?]@]BÏBÏDAEAG³G³I%J%L—L—N O Q{Q{SíSíU_V_XÑXÑZC[C]µ]µ_'`'b™b™d e g}g}iïiïkalanÓnÓpEqEs·s·u)v)x›x›z { }}ññ�c‚c„Õ„Õ†E‡E‰¶‰¶‹'Œ'Ž˜Ž˜� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1 ~!1� ~!1«¾!1 «¾!1MÅ!1MÅ!1Ùç!1€Ùç!1w("1üw("1/"1/"1«Q"1p«Q"1I’"1ìI’"1ñ˜"1ôñ˜"1ÍÚ"1`ÍÚ"1-ÿ"1Ì-ÿ"150#1ì50#1#1L #1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1·€&1T ·€&1‡R'1˜‡R'1/e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’B b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1•Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’B b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1•Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’B b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1•Õ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1„¤  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp å åZZÏÏ)D*D4¹4¹>.?.I£I£ST^Š^Šhühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppååZZÏÏ D D ¹ ¹..££��wwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇDG?GI¾I¾S3T3^¦^¦hisŠsŠ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssììeeÞÞ W W Ð ÐIIÂÂ;;´´--¦¦ ˜ ˜"#%Š%Š'(*|*|,õ,õ.n/n1ç1ç3`4`6Ù6Ù8R9R;Ë;Ë=D>D@½@½B6C6E¯E¯G(H(JŸJŸLMO�O�QRT{T{VòVòXiYi[à[à]W^W`Î`ÎbCcCe¸e¸g-h-j¢j¢lmoŠoŠqýqýsptpvãvãxVyV{É{É}<~<€¯€¯‚"ƒ"…•…•‡ˆŠxŠxŒêŒêŽ\�\‘ΑΓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1 ~!1� ~!1«¾!1 «¾!15Å!15Å!1ßç!1€ßç!1}("1ü}("1/"1/"1±Q"1p±Q"1O’"1ìO’"1Ù˜"1ôÙ˜"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1�&1T �&1çR'1˜çR'1�e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›I b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Á€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›I b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Á€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›I b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Á €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1nŽ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp ê êddÚÚ)P*P4Æ4Æ>:?:I®I®S"T"^–^–h i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppêêddÞÞ X X Ò ÒLLÆÆ@@ºº44®®(( ¢ ¢"#%’%’'(*~*~,ô,ô.j/j1à1à3V4V6Ì6Ì8B9B;¸;¸=.>.@¤@¤BCE�E�GHJzJzLîLîNbObQÖQÖSJTJV¾V¾X2Y2[¦[¦]^`Ž`Žbcevevgêgêi^j^lÒlÒnFoFqºqºs.t.v¢v¢xy{Š{Š}þ}þr€r‚æ‚æ„Z…Z‡·ΉBŠBŒ¶Œ¶Ž*�*‘ž‘ž“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ë}!1�Ë}!1i¾!1 i¾!1¿Â!1¿Â!1—ç!1€—ç!15("1ü5("1‘,"1‘,"1iQ"1piQ"1’"1ì’"1c–"1ôc–"1µÚ"1`µÚ"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1Õ€&1T Õ€&1¥R'1˜¥R'1Me'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ð b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ð b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ð b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1r’  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq è è__ÖÖ)M*M4Ä4Ä>9?9I®I®S!T!^’^’hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqèè__ÖÖ M M Ä Ä;;²²))  ŽŽ | |"ó"ó$j%j'á'á)X*X,Ï,Ï.F/F1½1½34446«6«8"9";™;™=>@‡@‡BþBþDuEuGìGìIaJaLÖLÖNKOKQÀQÀS5T5VªVªXY[”[”] ^ `~`~bñbñddedg×g×iJjJl½l½n.o.qŸqŸstv�v�xòxòzc{c}Ô}ÔE€E‚¶‚¶„'…'‡—‡—‰ŠŒyŒyŽêŽê�[‘[“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1 ~!1� ~!1«¾!1 «¾!15Å!15Å!1ßç!1€ßç!1}("1ü}("1/"1/"1±Q"1p±Q"1O’"1ìO’"1Ù˜"1ôÙ˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1/0#1ì/0#1û~#1L û~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›K b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Á€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›K b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Á€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›K b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Á €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1…¥  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp è è``ÔÔ)H*H4¼4¼>.?.I I ST^„^„höhörhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppèè``ØØ P P È È@@¸¸00¨¨  ˜˜ ˆ ˆ"#%t%t'è'è)\*\,Ð,Ð.D/D1¸1¸3,4,6 6 89;ˆ;ˆ=ü=ü?p@pBäBäDXEXGÌGÌI>J>L°L°N"O"Q”Q”STVxVxXêXêZ\[\]Î]Î_@`@b²b²d$e$g–g–ijlzlznìnìp^q^sÐsÐuBvBx´x´z&{&}˜}˜ € ‚|‚|„î„î†`‡`‰Ò‰Ò‹DŒD޶޶�(‘(“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Å}!1�Å}!1c¾!1 c¾!1­Â!1­Â!1—ç!1€—ç!15("1ü5("1,"1,"1iQ"1piQ"1’"1ì’"1Q–"1ôQ–"1—Ú"1`—Ú"1÷þ"1Ì÷þ"1)0#1ì)0#1õ~#1L õ~#1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1S�&1T S�&1#S'1˜#S'1Ëe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Ñ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Ñ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Ñ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Xx  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo ç ç__××)O*O4Ç4Ç>=?=I³I³S)T)^Ÿ^Ÿhis‹s‹} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooçç__×× O O Ç Ç??··//§§—— ‡ ‡"ÿ"ÿ$w%w'ï'ï)g*g,ß,ß.W/W1Ï1Ï3G4G6¿6¿8797;¯;¯='>'@Ÿ@ŸBCE�E�GHJ}J}LóLóNiOiQßQßSUTUVËVËXAYA[·[·]-^-`£`£bce�e�ghj{j{lñlñngogqÝqÝsStSvÉvÉx?y?{µ{µ}+~+€¡€¡‚ƒ…�…�‡ˆŠyŠyŒïŒïŽe�e‘Û‘Û“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1~!1�~!1±¾!1 ±¾!1MÅ!1MÅ!1åç!1€åç!1ƒ("1üƒ("1/"1/"1·Q"1p·Q"1U’"1ìU’"1ñ˜"1ôñ˜"1Ù"1`Ù"1eý"1Ìeý"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤R b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1žÞ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤R b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1žÞ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤R b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1žÞ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Åå  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii ß ßTTÉÉ)>*>4±4±>$?$I—I—S T ^}^}hðhðrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiißßUUËË A A · ·--££��yyîîc c"Ø"Ø$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=Ý=Ý?P@PBÃBÃD6E6G©G©IJL�L�NOQuQuSèSèU[V[XÎXÎZA[A]´]´_'`'bšbšd e g€g€ióiókflfnÙnÙpLqLs¿s¿u2v2x¥x¥z{}‹}‹þþ�q‚q„ä„ä†U‡U‰ljNj9Œ9ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1¿}!1�¿}!1]¾!1 ]¾!1­Â!1­Â!1‘ç!1€‘ç!1/("1ü/("1,"1,"1cQ"1pcQ"1’"1ì’"1Q–"1ôQ–"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–Í b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–Í b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–Í b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1s“  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ã ãYYÏÏ)E*E4»4»>/?/I£I£ST^‹^‹hÿhÿrsss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““  õõk k"á"á$W%W'Í'Í)C*C,¹,¹.///1¥1¥346‘6‘89;};}=ó=ó?i@iBßBßDUEUGËGËI?J?L³L³N'O'Q›Q›STVƒVƒX÷X÷Zk[k]ß]ß_S`SbÇbÇd;e;g¯g¯i#j#l—l—n o qqsósóugvgxÛxÛzO{O}Ã}Ã7€7‚«‚«„…‡“‡“‰ŠŒ{Œ{ŽïŽï�c‘c“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1~!1�~!1±¾!1 ±¾!1MÅ!1MÅ!1åç!1€åç!1ƒ("1üƒ("1/"1/"1·Q"1p·Q"1U’"1ìU’"1ñ˜"1ôñ˜"1Ù"1`Ù"1eý"1Ìeý"1)0#1ì)0#1õ~#1L õ~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%13€&1T 3€&1R'1˜R'1«d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤T  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ÿß€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤T  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ÿß€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤T  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ÿß €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Åå  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk á áTTÇÇ):*:4«4«>?I�I�SþSþ]o^ohàhàrOsO} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkááWWÍÍ C C ¹ ¹//¥¥‘‘wwêê] ]"Ð"Ð$C%C'¶'¶))*),œ,œ./1‚1‚3õ3õ5h6h8Û8Û:N;N=¿=¿?0@0B¡B¡DEGƒGƒIôIôKeLeNÖNÖPGQGS¸S¸U)V)XšXšZ [ ]|]|_í_ía^b^dÏdÏf@g@i±i±k"l"n“n“pqsusuuæuæwWxWzÈzÈ|9}9ªª�‚„Œ„Œ†û†ûˆk‰k‹Û‹Û�KŽK�»�»’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1¿}!1�¿}!1]¾!1 ]¾!1¿Â!1¿Â!1‘ç!1€‘ç!1/("1ü/("1‘,"1‘,"1cQ"1pcQ"1’"1ì’"1c–"1ôc–"1‹Ú"1`‹Ú"1ëþ"1Ìëþ"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Î b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Î b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Î b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ss  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk â âYYÐÐ)G*G4¾4¾>3?3I¨I¨ST^’^’his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkââYYÐÐ G G ¾ ¾55¬¬##ššˆˆÿÿv v"í"í$d%d'Û'Û)R*R,É,É.@/@1·1·3.4.6¥6¥89;“;“= > @�@�BøBøDoEoGæGæI[J[LÐLÐNEOEQºQºS/T/V¤V¤XY[Ž[Ž]^`x`xbíbídbebg×g×iLjLlÁlÁn6o6q«q«s t v•v•x y {{}ô}ôi€i‚Þ‚Þ„S…S‡ȇȉ=Š=Œ²Œ²Ž'�'‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1~!1�~!1·¾!1 ·¾!1/Å!1/Å!1åç!1€åç!1ƒ("1üƒ("1/"1/"1·Q"1p·Q"1U’"1ìU’"1Ó˜"1ôÓ˜"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Õ€&1T Õ€&1¥R'1˜¥R'1Me'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­[' b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‚€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­[' b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‚€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­[' b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‚ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Åå  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àUUÊÊ)?*?4´4´>(?(IœIœST^„^„høhørlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  ŠŠÿÿt t"é"é$^%^'Ó'Ó)H*H,½,½.2/21§1§346‘6‘89;{;{=ð=ð?e@eBÚBÚDOEOGÄGÄI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡Œ‡Œ‰ŠŒvŒvŽëŽë�`‘`“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1¹}!1�¹}!1W¾!1 W¾!1¹Â!1¹Â!1‹ç!1€‹ç!1)("1ü)("1‹,"1‹,"1]Q"1p]Q"1û‘"1ìû‘"1]–"1ô]–"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“þÊ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“þÊ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“þÊ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1t”  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj à àVVËË)@*@4µ4µ>(?(I›I›ST^�^�hôhôrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjààVVÌÌ B B ¸ ¸..¤¤��||òòh h"Þ"Þ$S%S'È'È)=*=,²,².'/'1œ1œ346†6†8û8û:p;p=å=å?Z@ZBÏBÏDDEDG¹G¹I,J,LŸLŸNOQ…Q…SøSøUkVkXÞXÞZQ[Q]Ä]Ä_7`7bªbªdeg�g�ijlvlvnénép\q\sÏsÏuBvBxµxµz({(}›}›€‚�‚�„ô„ô†g‡g‰Ú‰Ú‹MŒMŽÀŽÀ�3‘3“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1~!1�~!1·¾!1 ·¾!1/Å!1/Å!1ëç!1€ëç!1‰("1ü‰("1/"1/"1½Q"1p½Q"1[’"1ì[’"1Ó˜"1ôÓ˜"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1)0#1ì)0#1õ~#1L õ~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­]) b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1‚€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­]) b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1‚€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­]) b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1‚ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Åå  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞSSÈÈ)=*=4²4²>%?%I˜I˜S T ^|^|hîhîr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽxxííb b"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†P‡P‰ÉË6Œ6ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¹}!1�¹}!1W¾!1 W¾!1§Â!1§Â!1…ç!1€…ç!1#("1ü#("1y,"1y,"1WQ"1pWQ"1õ‘"1ìõ‘"1K–"1ôK–"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1á€&1T á€&1±R'1˜±R'1Ye'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ÿË b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ÿË b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ÿË b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Åå  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk å å[[ÑÑ)G*G4»4»>/?/I£I£ST^‹^‹hÿhÿrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkåå__ÙÙ S S Í ÍGGÁÁ;;µµ++¡¡ � �"#%y%y'ï'ï)e*e,Û,Û.Q/Q1Ç1Ç3=4=6³6³8)9);Ÿ;Ÿ=>@‡@‡BûBûDoEoGãGãIWJWLËLËN?O?Q³Q³S'T'V›V›XY[ƒ[ƒ]÷]÷_k`kbßbßdSeSgÇgÇi;j;l¯l¯n#o#q—q—s t vvxóxózg{g}Û}ÛO€O‚ÂÄ7…7‡©‡©‰ŠŒ�Œ�Ž�‘u‘u“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1~!1�~!1½¾!1 ½¾!1AÅ!1AÅ!1ëç!1€ëç!1‰("1ü‰("1/"1/"1½Q"1p½Q"1[’"1ì[’"1å˜"1ôå˜"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ó€&1T ó€&1ÃR'1˜ÃR'1ke'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶d0 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1šÚ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶d0 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1šÚ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶d0 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1šÚ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Åå  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I¥I¥ST^Ž^Žhisrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLÍLÍNDODQ»Q»S2T2V©V©X Y [—[—]^`…`…bübüdsesgêgêiajalØlØnJoJq¼q¼s.t.v v xy{„{„}ö}öh€h‚Ú‚Ú„L…L‡¾‡¾‰0Š0Œ¢Œ¢Ž�‘†‘†“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1³}!1�³}!1Q¾!1 Q¾!1§Â!1§Â!1…ç!1€…ç!1#("1ü#("1y,"1y,"1WQ"1pWQ"1õ‘"1ìõ‘"1K–"1ôK–"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1i€&1T i€&19R'1˜9R'1ád'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ûÇ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ûÇ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ûÇ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Åå  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áUUÉÉ)=*=4¯4¯>!?!I“I“ST^w^whéhérYsY} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‰‰ýýq q"å"å$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b�b�dÿdÿfqgqiãiãkUlUnÇnÇp9q9s«s«uvx�x�z{}s}såå�W‚W„ɄɆ9‡9‰ª‰ª‹ŒŽŒŽŒ�ý�ý’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1~!1�~!1½¾!1 ½¾!1AÅ!1AÅ!1ñç!1€ñç!1�("1ü�("1/"1/"1ÃQ"1pÃQ"1a’"1ìa’"1å˜"1ôå˜"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶f2 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1›Û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶f2 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1›Û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‚¶f2 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1›Û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Åå  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ù ÙNNÃÃ)8*84­4­>"?"I—I—S T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÙÙNNÃà 8 8 ­ ­""——  ��öökkààU U"Ê"Ê$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1³}!1�³}!1Q¾!1 Q¾!1³Â!1³Â!1ç!1€ç!1("1ü("1…,"1…,"1QQ"1pQQ"1ï‘"1ìï‘"1W–"1ôW–"1ùØ"1`ùØ"1Yý"1ÌYý"10#1ì0#1é~#1L é~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1/�&1T /�&1ÿR'1˜ÿR'1§e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�üÈ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�üÈ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�üÈ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Åå  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg à àYYÒÒ)K*K4Ä4Ä>;?;I²I²S'T'^œ^œhis€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggààYYÒÒ K K Ä Ä==¶¶//¨¨!!šš Œ Œ"#%~%~'÷'÷)p*p,é,é.b/b1Û1Û3T4T6Í6Í8F9F;¿;¿=8>8@±@±B*C*E£E£GHJ“J“L M O�O�QøQøSoToVæVæX]Y][Ô[Ô]K^K`Â`Âb7c7e¬e¬g!h!j–j–l m o€o€qóqósftfvÙvÙxLyL{¿{¿}2~2€¥€¥‚ƒ…‹…‹‡ü‡ü‰nŠnŒàŒàŽR�R‘đē b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1~!1�~!1½¾!1 ½¾!1#Å!1#Å!1ñç!1€ñç!1�("1ü�("1õ."1õ."1ÃQ"1pÃQ"1a’"1ìa’"1ǘ"1ôǘ"1ùØ"1`ùØ"1Yý"1ÌYý"10#1ì0#1é~#1L é~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‹¿m9 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Á€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‹¿m9 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Á€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‹¿m9 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Á €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Åå  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Þ ÞXXÎÎ)D*D4º4º>.?.I¢I¢ST^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÞÞXXÒÒ L L Æ Æ@@ºº44®®((¢¢ – –"#%†%†'ü'ü)r*r,è,è.^/^1Ô1Ô3J4J6À6À8696;¬;¬=">"@˜@˜BCE„E„GúGúInJnLâLâNVOVQÊQÊS>T>V²V²X&Y&[š[š]^`‚`‚böbödjejgÞgÞiRjRlÆlÆnDoDq¸q¸s,t,v v xy{ˆ{ˆ}ü}üp€p‚ä‚ä„X…X‡̇̉@Š@Œ´Œ´Ž(�(‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1­}!1�­}!1K¾!1 K¾!1³Â!1³Â!1ç!1€ç!1("1ü("1…,"1…,"1QQ"1pQQ"1ï‘"1ìï‘"1W–"1ôW–"1ùØ"1`ùØ"1Yý"1ÌYý"10#1ì0#1é~#1L é~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1½€&1T ½€&1�R'1˜�R'15e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�øÄ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�øÄ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�øÄ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Åå  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>-?-I¢I¢ST^Š^Šhûhûrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbåbådXeXgËgËi>j>l±l±n&o&q—q—stvyvyxêxêz[{[}Ì}Ì=€=‚®‚®„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1%~!1�%~!1þ!1 þ!1#Å!1#Å!1ñç!1€ñç!1�("1ü�("1õ."1õ."1ÃQ"1pÃQ"1a’"1ìa’"1ǘ"1ôǘ"1ùØ"1`ùØ"1Yý"1ÌYý"10#1ì0#1é~#1L é~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‹¿o; b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Á€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‹¿o; b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Á€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‹¿o; b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Á €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Åå  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜTTÈÈ)<*<4°4°>"?"I”I”ST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœŒŒ | |"ô"ô$h%h'Ü'Ü)P*P,Ä,Ä.8/81¬1¬3 4 6”6”89;|;|=ð=ð?d@dBØBØDLELGÀGÀI2J2L¤L¤NOQˆQˆSúSúUlVlXÞXÞZP[P]Â]Â_4`4b¦b¦degŠgŠiüiüknlnnènèpZqZsÌsÌu>v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1§}!1�§}!1E¾!1 E¾!1¡Â!1¡Â!1yç!1€yç!1("1ü("1s,"1s,"1KQ"1pKQ"1é‘"1ìé‘"1E–"1ôE–"1+Ú"1`+Ú"1‹þ"1Ô‹þ"10#1ô0#1é~#1T é~#1Û$1´ Û$1ñ$1x ñ$1áÀ%1| áÀ%1A�&1\ A�&17Q'1 7Q'1ßc'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒùÅ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒùÅ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒùÅ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1Ì\±  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSÍÍ)E*E4½4½>3?3I©I©ST^¡^¡his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=>@•@•B C E…E…GýGýIsJsLéLéN_O_QÕQÕSKTKVÁVÁX7Y7[­[­]#^#`™`™bce…e…gûgûiqjqlçlçnioiqßqßsUtUvËvËxAyA{·{·}-~-€£€£‚ƒ…�…�‡ˆŠ{Š{ŒñŒñŽg�g‘ݑݓ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!1%~!1�%~!1þ!1 þ!15Å!15Å!1÷ç!1€÷ç!1•("1ü•("1/"1/"1ÉQ"1pÉQ"1g’"1ìg’"1Ù˜"1ôÙ˜"1ÏØ"1`ÏØ"1/ý"1Ì/ý"10#1ì0#1é~#1L é~#1½$1¬ ½$1åð$1p åð$1ÃÀ%1t ÃÀ%1�€&1T �€&1ƒP'1˜ƒP'1+c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”ÈvB b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1˜Ø€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”ÈvB b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1˜Ø€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”ÈvB b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1˜Ø €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ÅÁ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10]] Ó ÓRRÇÇ)<*<4¯4¯>"?"I•I•ST^‡^‡húhúrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð]]ÓÓII¿¿ 5 5 « «!!——  ƒƒwwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Û=Û?N@NBÁBÁD4E4G§G§IJL�L�NOQsQsSæSæUYVYXÌXÌZ?[?]²]²_%`%b¤b¤degŠgŠiýiýkplpnãnãpVqVsÉsÉu'?'I›I›ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqƒqƒs÷s÷ukvkxßxßzS{S}Ç}Ç;€;‚¯‚¯„#…#‡—‡—‰ Š ŒŒŽóŽó�g‘g“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!1+~!1�+~!1ɾ!1 ɾ!1;Å!1;Å!1÷ç!1€÷ç!1•("1ü•("1 /"1 /"1ÉQ"1pÉQ"1g’"1ìg’"1ߘ"1ôߘ"1ÏØ"1`ÏØ"1/ý"1Ì/ý"10#1ì0#1é~#1L é~#1¥$1¬ ¥$1Íð$1p Íð$1«À%1t «À%1'€&1T '€&1P'1˜P'1Åb'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”ÈxD b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1™Ù€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”ÈxD b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1™Ù€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1”ÈxD b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1™Ù €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ÅÁ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Õ ÕPPÃÃ)6*64§4§>?I‰I‰ST^w^whèhèrWsW} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……ssææY Y"Ì"Ì$?%?'²'²)%*%,˜,˜. / 1~1~3ñ3ñ5d6d8×8×:J;J=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éafbfd×d×fHgHi¹i¹k*l*n›n›p q s}s}uîuîw_x_zÐzÐ|A}A²²�#‚#„”„”†‡‰s‰s‹ã‹ã�SŽS�Ã�Ã’ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„1i!1e}!1e}!1¾!1|¾!1gÃ!1„gÃ!1�Æ!1ð�Æ!1'Ð!1|'Ð!1]Ó!1H]Ó!11ç!1Ô1ç!1Ï'"1PÏ'"19-"1X9-"1o0"1Äo0"1ù9"1Pù9"1/="1/="1Q"1¨Q"1¡‘"1$ ¡‘"1 —"1,  —"1Aš"1˜ Aš"1Ë£"1$ Ë£"1§"1ð §"1CÔ"1| CÔ"1 Ú"1X Ú"1mþ"1Ômþ"1¯+#1ô¯+#1y:#1Ðy:#15o#1\5o#1µŽ#1€µŽ#1³™#1¬³™#1Ï$18Ï$1§ò$1ü§ò$1áÞ%1áÞ%1[ì%1à[ì%1�&1l�&1ÕR'1°ÕR'1}e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¿‹ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ç bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¿‹ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ç  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1P¿‹  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ç bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ççm�ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1W—î.…ÅP�\  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dW×.®…Ü\³3Š aá8¸�æfæf¿?½= ˜ ”  q ñ k ë B   ™ ð p Ç ^žõ5à ŒÌ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Û ÛÕÕ)K*K4Á4Á>7?7I«I«ST^Ÿ^Ÿhis‡s‡} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÛÛUUÏÏ I I à Ã==··11««ëëeeßßY Y"Ó"Ó$�%�'(*y*y,ï,ï.e/e1Û1Û3Q4Q6Ç6Ç8=9=;³;³=)>)@Ÿ@ŸBCE‹E‹GHJwJwLëLëN_O_QÓQÓSGTGV»V»X/Y/[£[£]^`‹`‹bÿbÿdsesgçgçi[j[lÏlÏnOoOqÃqÃs7t7v«v«xy{“{“}~€{€{‚ï‚ï„c…c‡ׇ׉KŠKŒ¿Œ¿Ž3�3‘§‘§“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1=~!1è=~!1Û¾!1dÛ¾!1‰Å!1l‰Å!1Ï!1èÏ!1IÒ!1´IÒ!1è!1@è!1­("1¼­("1[/"1Ä[/"1å8"1@å8"1<"1 <"1áQ"1˜áQ"1’"1’"1-™"1 -™"1·¢"1˜ ·¢"1í¥"1d í¥"1/Ó"1ð /Ó"1—Ú"1Ì —Ú"1÷þ"1H ÷þ"19,#1h9,#1y:#1Dy:#15o#1Ð5o#1µŽ#1ôµŽ#1³™#1 ³™#1Õ$1¬Õ$1­ò$1p­ò$1çÞ%1tçÞ%1aì%1Taì%1/�&1à/�&1ÿR'1$ÿR'1§e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1°ä”` b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—q±î. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1°ä”` b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—q±î. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1°ä”`  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—q±î.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‰©Bbr’ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1W—î.B‚…Å\  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dW×.®…Ü\³3Š aá8¸À@�•æfjê½= ? ¿ ”  k ë B   ™ ð p Ç ^ž’Òõ5ŒÌ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ý ÝVV——)J*J4Ã4Ã>7@°@°B)C)E¢E¢GHJ”J”L M O‚O‚QùQùSpTpVçVçX^Y^[Õ[Õ]L^L`Ã`Ãb8c8e­e­g"h"j—j—l m o‹o‹qþqþsqtqväväxWyW{Ê{Ê}=~=€°€°‚#ƒ#…–…–‡ ˆ Š{Š{ŒíŒíŽ_�_‘ёѓ  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„1i!1_}!1_}!1ý½!1|ý½!1gÃ!1„gÃ!1�Æ!1ð�Æ!1'Ð!1|'Ð!1]Ó!1H]Ó!11ç!1Ô1ç!1Ï'"1PÏ'"19-"1X9-"1o0"1Äo0"1ù9"1Pù9"1/="1/="1Q"1¨Q"1¡‘"1$ ¡‘"1 —"1,  —"1Aš"1˜ Aš"1Ë£"1$ Ë£"1§"1ð §"1CÔ"1| CÔ"11Ú"1X1Ú"1‘þ"1Ô‘þ"1Ó+#1ôÓ+#1y:#1Ðy:#15o#1\5o#1µŽ#1€µŽ#1³™#1¬³™#1·$18·$1�ò$1ü�ò$1ÉÞ%1ÉÞ%1Cì%1àCì%1�€&1l�€&1]R'1°]R'1e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¼ˆ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ç bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¼ˆ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ç  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¼ˆ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ç bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÇçnŽ , bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1W—î.…Å\ é)  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dW×.®…Ü\³3Š aá8¸�æf½= ”  k ë J Ê B  # £  ™ ü | ð p Õ U Ç ^žõ5ŒÌy¹ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ù ÙÍÍ)A*A4µ4µ>)?)I›I›S T ^‹^‹hýhýroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A ¹ ¹11©©!!™™ÖÖNNÆÆ> >"¶"¶$m%m'á'á)U*U,É,É.=/=1±1±3%4%6™6™8 9 ;�;�=õ=õ?i@iBÝBÝDQEQGÅGÅI9J9L«L«NOQ�Q�STVsVsXåXåZW[W]É]É_;`;b­b­deg‘g‘ijlulunónópeqes×s×uIvIx»x»z-{-}Ÿ}Ÿ€‚ƒ‚ƒ„õ„õ†g‡g‰ىًKŒK޽޽�/‘/“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1C~!1èC~!1á¾!1dá¾!1‰Å!1l‰Å!1Ï!1èÏ!1IÒ!1´IÒ!1è!1@è!1­("1¼­("1[/"1Ä[/"1å8"1@å8"1<"1 <"1áQ"1˜áQ"1’"1’"1-™"1 -™"1·¢"1˜ ·¢"1í¥"1d í¥"1/Ó"1ð /Ó"1gÚ"1Ì gÚ"1Çþ"1H Çþ"1 ,#1h ,#1¯:#1D¯:#1ko#1Ðko#1½�#1ô½�#1»š#1 »š#1Ã$1¬Ã$1›ò$1p›ò$1ÕÞ%1tÕÞ%1Oì%1TOì%1½€&1ཀ&1�R'1$�R'15e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1²æ˜d b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—r²î. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1²æ˜d b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—r²î. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1²æ˜d  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—r²î.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 y™Aa% bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1e¥ J­í× P�  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1deåJÊ/¯”ùyÞ^ÃC¨(‹ nîQ Ñ 4 ´  ”  — é i ú z ¾ > Ý ] “  À @¥åJŠ'gí-�Ð bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPP��)B*B4¹4¹>0?0I¥I¥ST^•^•hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$$%$'›'›)*,‰,‰./1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n1o1q¢q¢stv„v„xõxõzf{f}×}×H€H‚¹‚¹„*…*‡œ‡œ‰ Š Œ~Œ~ŽïŽï�`‘`“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1§}!1è§}!1E¾!1dE¾!1�Å!1l�Å!1Ï!1èÏ!1OÒ!1´OÒ!1yç!1@yç!1("1¼("1a/"1Äa/"1ë8"1@ë8"1!<"1 !<"1KQ"1˜KQ"1é‘"1é‘"13™"1 3™"1½¢"1˜ ½¢"1ó¥"1d ó¥"15Ó"1ð 5Ó"1mÚ"1Ì mÚ"1Íþ"1H Íþ"1,#1h,#1÷:#1D÷:#1³o#1гo#1‘#1ô‘#1œ#1 œ#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1�€&1à�€&1QR'1$QR'1ùd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹úÆ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X˜ï/ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹úÆ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X˜ï/ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹úÆ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X˜ï/  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 nŽDdñ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1t´(hW—Þ”Ô  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dtôhè\ÜPÐDÄ8¸,¬  ùy–ÔT¯/  Œ Œ  ‚ ø x î n ä d Ú ZÐP IÄx¸.nä$ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ Ô ÔII¾¾)3*34¨4¨>?ITITS T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÔÔJJÀÀ 6 6 ¬ ¬""˜˜„„ùùnnããX X"Í"Í$B%B'·'·),*,,¡,¡./1‹1‹346u6u8ê8ê:_;_=Ô=Ô?G@GBºBºD-E-G G IJL†L†NùNùPlQlSßSßUVX‹X‹ZþZþ\q]q_ä_äa›b›deg�g�iôiôkglgnÚnÚpMqMsÀsÀu3v3x¦x¦z{}Œ}Œÿÿ�r‚r„å„å†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!17~!1è7~!1Õ¾!1dÕ¾!1�Å!1l�Å!1Ï!1èÏ!1OÒ!1´OÒ!1 è!1@ è!1§("1¼§("1a/"1Äa/"1ë8"1@ë8"1!<"1 !<"1ÛQ"1˜ÛQ"1y’"1y’"13™"1 3™"1½¢"1˜ ½¢"1ó¥"1d ó¥"15Ó"1ð 5Ó"1»Ú"1Ì »Ú"1ÿ"1H ÿ"1],#1h],#1:#1D:#1;o#1Ð;o#1»Ž#1ô»Ž#1¹™#1 ¹™#1Ï$1¬Ï$1§ò$1p§ò$1áÞ%1táÞ%1[ì%1T[ì%15�&1à5�&1S'1$S'1­e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¥Ù‰U b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—©éî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¥Ù‰U b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—©éî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¥Ù‰U  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—©éî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ˜¸@`�¯ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1W—î.=}…Å\  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dW×.®…Ü\³3Š aá8¸ÇG�žæfuõ½= L Ì ”  k ë B   ™ ð p Ç ^ž­íõ5ŒÌ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜTTÌÌ)D*D4¼4¼>4?4IªIªS T ^^^^his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœŒŒ | |"ô"ô$l%l'ä'ä)\*\,Ô,Ô.L/L1Ä1Ä3<4<6´6´8,9,;¤;¤=>@”@”B C E„E„GüGüItJtLêLêN`O`QÖQÖSLTLVÂVÂX8Y8[®[®]$^$`š`šbce†e†gügüirjrlèlèn&o&qœqœstvˆvˆxþxþz®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1§}!1è§}!1E¾!1dE¾!1�Å!1l�Å!1Ï!1èÏ!1OÒ!1´OÒ!1sç!1@sç!1("1¼("1a/"1Äa/"1ë8"1@ë8"1!<"1 !<"1EQ"1˜EQ"1ã‘"1ã‘"13™"1 3™"1½¢"1˜ ½¢"1ó¥"1d ó¥"15Ó"1ð 5Ó"1CÚ"1Ì CÚ"1£þ"1H £þ"1å+#1hå+#1Ó:#1DÓ:#1�o#1Ð�o#1_�#1ô_�#1]›#1 ]›#1Ÿ$1¬Ÿ$1wò$1pwò$1±Þ%1t±Þ%1+ì%1T+ì%1!€&1à!€&1ñQ'1$ñQ'1™d'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹øÄ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X˜ï/ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹øÄ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X˜ï/ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹øÄ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X˜ï/  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 s“78 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1m­ZÅò2 p°  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dmíZÚGÇ4´!¡Žû{èhÓS¾> © ) ”  g ç  ÿ B   � j ê ø x U Õ @À-mÚ¤ä…Å0p bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ö ÖII¼¼)/*/4¢4¢>?IIIISýSý]n^nhßhßrPsP} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††ùùllßßR R"Å"Å$8%8'«'«)*,‘,‘./1w1w3ê3ê5]6]8Ð8Ð:C;C=¶=¶?'@'B˜B˜D E GzGzIëIëK\L\NÍNÍP>Q>S¯S¯UåUåWVXVZÇZÇ\8]8_©_©a]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ü†üˆl‰l‹Ü‹Ü�LŽL�¼�¼’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1=~!1è=~!1Û¾!1dÛ¾!1�Å!1l�Å!1Ï!1èÏ!1OÒ!1´OÒ!1 è!1@ è!1§("1¼§("1a/"1Äa/"1ë8"1@ë8"1!<"1 !<"1ÛQ"1˜ÛQ"1y’"1y’"13™"1 3™"1½¢"1˜ ½¢"1ó¥"1d ó¥"15Ó"1ð 5Ó"1‘Ú"1Ì ‘Ú"1ñþ"1H ñþ"13,#1h3,#1:#1D:#1;o#1Ð;o#1»Ž#1ô»Ž#1¹™#1 ¹™#1½$1¬½$1•ò$1p•ò$1ÏÞ%1tÏÞ%1Iì%1TIì%1À&1àÀ&1“R'1$“R'1;e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1§Û�Y b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—¨èî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1§Û�Y b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—¨èî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1§Û�Y  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—¨èî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x˜'f† bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1W—î.…ÅÔ \  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dW×.®…Ü\³3Š aá8¸�æf½= ”  # £ k ë ú z B Â Ñ Q  ™ ¨ ( ð p Ç ^žõ5D„ŒÌ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÄÄ):*:4°4°>&?&IšIšST^K^Khúhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$L%L'Â'Â)8*8,®,®.$/$1š1š346†6†8ü8ü:r;r=è=è?^@^BÔBÔDJEJGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnËnËp?q?s³s³u'v'x›x›zJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1­}!1Э}!1K¾!1LK¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ç!1(ç!1("1¤("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1QQ"1€QQ"1ï‘"1üï‘"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1µ7#1,µ7#1UŽ#1˜UŽ#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒûÇ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UW—î. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒûÇ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UW—î. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒûÇ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UW—î.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 o�÷a� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1T”€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1‚ß_¼< ™  $d €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ù ÙNNÃÃ)8*84­4­>#?#I—I—S T ^^hóhórisi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A ¹ ¹11©©!!™™ƒƒøøm m"â"â$W%W'Ì'Ì)A*A,¶,¶.+/+1 1 346Š6Š8ÿ8ÿ:t;t=é=é?^@^BÓBÓDHEHG½G½I3J3L§L§NOQ�Q�STVwVwXëXëZ_[_]Ó]Ó_G`Gb»b»d/e/g£g£ijl‹l‹nÿnÿpsqssçsçu[v[xÏxÏzC{C}·}·+€+‚Ÿ‚Ÿ„…‡‰‡‰‰þ‰þ‹sŒsŽèŽè�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!11~!1Ð1~!1Ͼ!1LϾ!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1ýç!1(ýç!1›("1¤›("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1ÏQ"1€ÏQ"1m’"1üm’"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1ÇÚ"1´ ÇÚ"1'ÿ"10 'ÿ"1i,#1Pi,#1�7#1,�7#1õ�#1˜õ�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ÿ€&1Ìÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎ~J b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—¿ò2 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎ~J b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—¿ò2 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎ~J  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—¿ò2  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 —·>^–¶ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations18x€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÎN§'€ Y Ù È €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I£I£ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbèbèd]e]gÒgÒiGjGl¼l¼n1o1q¦q¦stv�v�xy{z{z}ï}ïd€d‚قلN…N‡ÇÉ8Š8Œ­Œ­Ž"�"‘—‘—“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1­}!1Э}!1K¾!1LK¾!1•Å!1T•Å!1Ï!1ÐÏ!1UÒ!1œUÒ!1yç!1(yç!1("1¤("1g/"1¬g/"1ñ8"1(ñ8"1'<"1ô'<"1KQ"1€KQ"1é‘"1üé‘"19™"1 9™"1â"1€ â"1ù¥"1L ù¥"1;Ó"1Ø ;Ó"1mÚ"1´ mÚ"1Íþ"10 Íþ"1,#1P,#1U:#1,U:#1ñŽ#1˜ñŽ#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒùÅ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 VW—î. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒùÅ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 VW—î. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒùÅ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 VW—î.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x˜<\ @ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ñ1 €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1v ö S Ó 0 ° � Á €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×LLÁÁ)6*64«4«> ? I“I“ST^w^whéhér]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡üüqqææ[ ["Ð"Ð$E%E'º'º)/*/,¤,¤./1Ž1Ž346x6x8í8í:b;b=×=×?L@LBÁBÁD6E6G«G«I J L“L“NOQyQySìSìU_V_XÒXÒZE[E]¸]¸_+`+b�b�deg�g�ióiókelen×n×pIqIs»s»u-v-xŸxŸz{}ƒ}ƒõõ�g‚g„لنM‡M‰À‰À‹3Œ3ަަ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!11~!1Ð1~!1Ͼ!1LϾ!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1è!1(è!1¡("1¤¡("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1ÕQ"1€ÕQ"1s’"1üs’"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1©Ú"1´ ©Ú"1 ÿ"10  ÿ"1K,#1PK,#11:#1,1:#1‘Ž#1˜‘Ž#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐ‚N b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—¿ó3 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐ‚N b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—¿ó3 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐ‚N  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—¿ó3  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w— )y™ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ñ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations12 ² ‹ ä d ½ = a¡ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙOOÄÄ)9*94®4®>#?#I–I–S T ^|^|hïhïrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰ÿÿuuëëa a"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnänäpWqWsÊsÊu=v=x°x°z#{#}–}– € ‚|‚|„ï„ï†b‡b‰Õ‰Õ‹HŒH޻޻�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1³}!1г}!1Q¾!1LQ¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1ç!1(ç!1("1¤("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1QQ"1€QQ"1ï‘"1üï‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1¯Ú"1´ ¯Ú"1ÿ"10 ÿ"1Q,#1PQ,#1µ7#1,µ7#1sŽ#1˜sŽ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1ó€&1Ìó€&1ÃR'1ÃR'1ke'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�üÈ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Y™ð0 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�üÈ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Y™ð0 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�üÈ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Y™ð0  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p�ý~ž bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1O�€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ‰èhÇG ¦ & ? €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>-?-I¤I¤ST^�^�hÿhÿrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜIUJULÌLÌNCOCQºQºS1T1V¨V¨XY[–[–] ^ `„`„bûbûdrergégéi`j`l×l×nIoIq»q»s-t-vŸvŸxy{ƒ{ƒ}õ}õg€g‚قلK…K‡½‡½‰/Š/Œ¡Œ¡Ž�‘…‘…“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1+~!1Ð+~!1ɾ!1Lɾ!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1÷ç!1(÷ç!1•("1¤•("1y/"1¬y/"19"1(9"19<"1ô9<"1ÉQ"1€ÉQ"1g’"1üg’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1aÚ"1´ aÚ"1Áþ"10 Áþ"1,#1P,#1©7#1,©7#1aŽ#1˜aŽ#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1á€&1Ìá€&1±R'1±R'1Ye'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ãs? b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—žÞý= bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ãs? b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—žÞý= bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ãs?  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—žÞý=  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ’²Çç4T bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€Þ^¼< š  0p €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh â âXXÎÎ)D*D4º4º>.?.I¢I¢ST^Š^Šhþhþrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²((žž Š Š"#%v%v'ì'ì)b*b,Ø,Ø.N/N1Ä1Ä3:4:6°6°8&9&;œ;œ=>@†@†BúBúDnEnGâGâIVJVLÊLÊN>O>Q²Q²S&T&VšVšXY[‚[‚]ö]ö_j`jbÞbÞdReRgÆgÆi:j:l®l®n"o"q–q–s t v~v~xòxòzf{f}Ú}ÚN€N‚‚„6…6‡ª‡ª‰ŠŒ�Œ�Ž�‘v‘v“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1­}!1Э}!1K¾!1LK¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1ç!1(ç!1("1¤("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1QQ"1€QQ"1ï‘"1üï‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1‘Ú"1´ ‘Ú"1ñþ"10 ñþ"13,#1P3,#1U:#1,U:#1�#1˜�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�úÆ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X˜ï/ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�úÆ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X˜ï/ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�úÆ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X˜ï/  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 }�_*J bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1î. €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1…  d ä C à " ¢ Þ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ö ÖKKÀÀ)5*54ª4ª>!?!I–I–S T ^}^}hïhïrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÖÖKKÀÀ 5 5 ª ª””  ~~óóhhÝÝR R"Ç"Ç$<%<'±'±)&*&,›,›./1…1…3ú3ú5o6o8ä8ä:Y;Y=Î=Î?C@CB¸B¸D-E-G¢G¢IJLŽLŽNOQxQxSíSíUbVbX×X×ZL[L]Á]Á_6`6b«b«d e g•g•i j llnñnñpcqcsÕsÕuGvGx¹x¹z+{+}�}�€‚�‚�„ó„ó†e‡e‰׉׋IŒI޻޻�-‘-“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1+~!1Ð+~!1ɾ!1Lɾ!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1ýç!1(ýç!1›("1¤›("1y/"1¬y/"19"1(9"19<"1ô9<"1ÏQ"1€ÏQ"1m’"1üm’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"17Ú"1´ 7Ú"1—þ"10 —þ"1Ù+#1PÙ+#1C:#1,C:#1�#1˜�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1i€&1Ìi€&19R'19R'1ád'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘ÅwC b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—�Ýþ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘ÅwC b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—�Ýþ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘ÅwC  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—�Ýþ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 r’Çç' bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Þ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1x ø V Ö 4 ´  ’ Î €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞRRÆÆ):*:4®4®> ? I’I’ST^v^vhèhèrZsZ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žž††úún n"â"â$V%V'Ê'Ê)>*>,²,².&/&1š1š346‚6‚8ö8ö:j;j=Þ=Þ?P@PBÂBÂD4E4G¦G¦IJLŠLŠNüNüPnQnSàSàURVRXÄXÄZ6[6]¨]¨_`bŒbŒdþdþfpgpiâiâkTlTnÆnÆp8q8sªsªuvxŽxŽz{}r}rää�V‚V„ȄȆ:‡:‰«‰«‹ŒŽ�Ž��þ�þ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¹}!1й}!1W¾!1LW¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1…ç!1(…ç!1#("1¤#("1…/"1¬…/"19"1(9"1E<"1ôE<"1WQ"1€WQ"1õ‘"1üõ‘"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1¯Ú"1´ ¯Ú"1ÿ"10 ÿ"1Q,#1PQ,#1µ7#1,µ7#1‘Ž#1˜‘Ž#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1�&1Ì�&1ÕR'1ÕR'1}e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽýÉ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Io¯F bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽýÉ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Io¯F bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽýÉ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Io¯F  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 q‘!w— bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1K‹€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1‘òrÓS ´ 4 [› €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee ß ßYYÏÏ)E*E4»4»>1?1I¥I¥ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯))££ — —"#%‡%‡'ý'ý)s*s,é,é._/_1Õ1Õ3K4K6Á6Á8797;­;­=#>#@™@™BCE…E…GûGûIqJqLåLåNYOYQÍQÍSATAVµVµX)Y)[�[�]^`…`…bùbùdmemgágáiUjUlÉlÉn=o=q±q±s%t%v™v™x y {�{�}õ}õi€i‚݂݄Q…Q‡Ňʼn9Š9Œ­Œ­Ž!�!‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1~!1Ð~!1½¾!1L½¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1ñç!1(ñç!1�("1¤�("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1ÃQ"1€ÃQ"1a’"1üa’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#1Á7#1,Á7#1-�#1˜-�#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%15�&1Ì5�&1S'1S'1­e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸h4 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—¿ñ1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸h4 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—¿ñ1 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸h4  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—¿ñ1  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Çç…¥: bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1V–€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1XØAÁ* ª  “ ö6 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh á áZZÓÓ)L*L4Å4Å>>?>IµIµS*T*^�^�hisƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhááZZÓÓ L L Å Å>>··00©©""›› � �"#%%'ø'ø)q*q,ê,ê.c/c1Ü1Ü3U4U6Î6Î8G9G;À;À=9>9@²@²B+C+E¤E¤GHJ–J–L M O„O„QûQûSrTrVéVéX`Y`[×[×]N^N`Å`Åb:c:e¯e¯g$h$j™j™lmo�o�qôqôsgtgvÚvÚxMyM{À{À}3~3€¦€¦‚ƒ…Œ…Œ‡ÿ‡ÿ‰qŠqŒãŒãŽU�U‘ǑǓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1³}!1г}!1Q¾!1LQ¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1…ç!1(…ç!1#("1¤#("1…/"1¬…/"19"1(9"1E<"1ôE<"1WQ"1€WQ"1õ‘"1üõ‘"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1�Ú"1´ �Ú"1ýþ"10 ýþ"1?,#1P?,#1U:#1,U:#13�#1˜3�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽûÇ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jn®E bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽûÇ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jn®E bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽûÇ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jn®E  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 € 0Phˆ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ì, €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1•  v ö W × 8 ¸ ü< €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝUUÉÉ)=*=4±4±>%?%I—I—S T ^{^{híhír_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%���� } }"õ"õ$i%i'Ý'Ý)Q*Q,Å,Å.9/91­1­3!4!6•6•8 9 ;};}=ñ=ñ?e@eBÙBÙDMEMGÁGÁI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnãnãpUqUsÇsÇu9v9x«x«z{}�}�€‚s‚s„å„å†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1%~!1Ð%~!1þ!1Lþ!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1÷ç!1(÷ç!1•("1¤•("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1ÉQ"1€ÉQ"1g’"1üg’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1%Ú"1´ %Ú"1…þ"10 …þ"1Ç+#1PÇ+#1g:#1,g:#1Õ�#1˜Õ�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºl8 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—¿ò2 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºl8 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—¿ò2 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºl8  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—¿ò2  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 e…Ççõ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ÿ? €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ü | å e Î N · 7Ÿß €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ý ÝTTËË)B*B4¹4¹>0?0I¥I¥ST^‰^‰húhúrlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÝÝTTËË B B ¹ ¹00§§••  ƒƒúúq q"è"è$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n%o%q–q–stvxvxxéxézZ{Z}Ë}Ë<€<‚­‚­„…‡�‡�‰ŠŒrŒrŽãŽã�T‘T“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1¹}!1й}!1W¾!1LW¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1‹ç!1(‹ç!1)("1¤)("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1]Q"1€]Q"1û‘"1üû‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1%Ú"1´ %Ú"1…þ"10 …þ"1Ç+#1PÇ+#1µ7#1,µ7#1‘Ž#1˜‘Ž#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1�€&1Ì�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�þÊ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X˜ï/ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�þÊ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X˜ï/ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�þÊ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X˜ï/  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ççhˆñ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D„€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ŠëkÌL ­ - T” €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØMMÂÂ)7*74¬4¬>?I’I’ST^x^xhëhër^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?K@KB¾B¾D1E1G¤G¤IJLŠLŠNýNýPpQpSãSãUVVVXÉXÉZ<[<]¯]¯_"`"b•b•deg{g{iîiîkalanÔnÔpGqGsºsºu-v-x x z{}†}†ùù�l‚l„߄߆R‡R‰ĉċ6Œ6ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1~!1Ð~!1·¾!1L·¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1ëç!1(ëç!1‰("1¤‰("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1½Q"1€½Q"1[’"1ü[’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"17Ú"1´ 7Ú"1—þ"10 —þ"1Ù+#1PÙ+#1Ù7#1,Ù7#1ù�#1˜ù�#1á$1$á$1¹ò$1è¹ò$1óÞ%1ìóÞ%1G�&1ÌG�&1S'1S'1¿e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­]) b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W——×ï/ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­]) b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W——×ï/ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1y­])  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W——×ï/  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ççx˜  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1l¬€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1°0¤$ ˜  Œ ¼ü €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àXXÐÐ)H*H4À4À>8?8I®I®S$T$^š^šhis†s†} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  �� € €"ø"ø$p%p'è'è)`*`,Ø,Ø.P/P1È1È3@4@6¸6¸8090;¨;¨= > @˜@˜BCEˆEˆGHJxJxLîLîNdOdQÚQÚSPTPVÆVÆX?IˆIˆSùSù]j^jhÛhÛrLsL} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠýýppããV V"É"É$<%<'¯'¯)"*",•,•./1{1{3î3î5a6a8Ô8Ô:G;G=º=º?+@+BœBœD E G~G~IïIïK`L`NÑNÑPBQBS³S³U$V$X•X•Z[]w]w_è_èaYbYdÊdÊf;g;i¬i¬klnŽnŽpÿpÿrpspuáuáwRxRzÃzÃ|4}4¥¥�‚„‡„‡†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1~!1Ð~!1½¾!1L½¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1ñç!1(ñç!1�("1¤�("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1ÃQ"1€ÃQ"1a’"1üa’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1Ú"1´ Ú"1sþ"10 sþ"1µ+#1Pµ+#1‘:#1,‘:#1­�#1˜­�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯a- b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W——×ï/ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯a- b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W——×ï/ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯a-  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W——×ï/  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÇçXxßÿ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 ` €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ t ô h è \Üp° €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜRRÈÈ)>*>4´4´>*?*IžIžST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒxxîîd d"Ú"Ú$P%P'Æ'Æ)<*<,²,².(/(1ž1ž346Š6Š89;v;v=ì=ì?b@bBØBØDNENGÄGÄI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1¿}!1п}!1]¾!1L]¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1‘ç!1(‘ç!1/("1¤/("1y/"1¬y/"19"1(9"19<"1ô9<"1cQ"1€cQ"1’"1ü’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1Ú"1´ Ú"1yþ"10 yþ"1»+#1P»+#1Ç7#1,Ç7#13�#1˜3�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1Û€&1ÌÛ€&1«R'1«R'1Se'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ÿË b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X˜ï/ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ÿË b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X˜ï/ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ÿË  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X˜ï/  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Çç]}ßÿ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1R’€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1HÈ3³ ž ‰ ò2 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝRRÇÇ)<*<4±4±>'?'I›I›ST^ƒ^ƒh÷h÷rmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��‡‡üüq q"æ"æ$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡�‡�‰ŠŒwŒwŽìŽì�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1~!1Ð~!1±¾!1L±¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1åç!1(åç!1ƒ("1¤ƒ("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1·Q"1€·Q"1U’"1üU’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1Ú"1´ Ú"1þ"10 þ"1Á+#1PÁ+#1÷7#1,÷7#1¿�#1˜¿�#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1�&1Ì�&1áR'1áR'1‰e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢R b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—�Áî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢R b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—�Áî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢R  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—�Áî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ççk‹æ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1‚€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ˆ ‡  †  … ‚ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>2?2I§I§ST^‘^‘his{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbìbìdaeagÖgÖiKjKlÀlÀn5o5qªqªstv”v”x y {~{~}ó}óh€h‚݂݄R…R‡LJlj<Š<Œ±Œ±Ž&�&‘›‘›“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1¿}!1п}!1]¾!1L]¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1‘ç!1(‘ç!1/("1¤/("1y/"1¬y/"19"1(9"19<"1ô9<"1cQ"1€cQ"1’"1ü’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#1m:#1,m:#1Û�#1˜Û�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ýÉ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X˜ï/ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ýÉ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X˜ï/ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ýÉ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X˜ï/  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ççr’4 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ý= €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ô t ß _ Ê J µ 5�Ý €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛPPÅÅ):*:4¯4¯>$?$I—I—S T ^{^{híhírasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹uuêê_ _"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I$J$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b¡b¡deg…g…i÷i÷kilinÛnÛpMqMs¿s¿u1v1x£x£z{}‡}‡ùù�k‚k„݄݆Q‡Q‰ĉċ7Œ7ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1~!1Ð~!1·¾!1L·¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1åç!1(åç!1ƒ("1¤ƒ("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1·Q"1€·Q"1U’"1üU’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1Ú"1´ Ú"1aþ"10 aþ"1£+#1P£+#1µ:#1,µ:#1‘#1˜‘#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1«€&1Ì«€&1{R'1{R'1#e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤V" b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W—€Àî. bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤V" b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W—€Àî. bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤V"  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W—€Àî.  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÇçKkÉé bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1A� €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 „  ƒ ‚�A� €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝSSÈÈ)=*=4²4²>'?'IšIšS T ^€^€hóhórfsf} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��yyïïe e"Û"Û$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I+J+LžLžNOQ„Q„S÷S÷UjVjXÝXÝZP[P]Ã]Ã_6`6b©b©deg�g�ijlulunènèp[q[sÎsÎuAvAx´x´z'{'}š}š € ‚€‚€„ó„ó†f‡f‰ىًLŒLŽ¿Ž¿�2‘2“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1Å}!1ÐÅ}!1c¾!1Lc¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1—ç!1(—ç!15("1¤5("1y/"1¬y/"19"1(9"19<"1ô9<"1iQ"1€iQ"1’"1ü’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1 Ú"1´  Ú"1mþ"10 mþ"1¯+#1P¯+#1ß7#1,ß7#1ù�#1˜ù�#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1�&1Ì�&1ÕR'1ÕR'1}e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ì b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X˜ï/ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ì b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X˜ï/ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ì  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X˜ï/  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÇçRrÍí bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1h¨€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  – Œ ‚  ¸ø €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>1?1I¨I¨ST^‘^‘hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIYJYLÐLÐNGOGQ¾Q¾S5T5V¬V¬X#Y#[š[š]^`ˆ`ˆbÿbÿdvevgígíidjdlÛlÛnMoMq¿q¿s1t1v£v£xy{‡{‡}ù}ùk€k‚݂݄O…O‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1 ~!1Ð ~!1«¾!1L«¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1ßç!1(ßç!1}("1¤}("1y/"1¬y/"19"1(9"19<"1ô9<"1±Q"1€±Q"1O’"1üO’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1 8#1, 8#1g‘#1˜g‘#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1ó€&1Ìó€&1ÃR'1ÃR'1ke'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—G b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W——×ï/ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—G b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W——×ï/ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—G  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W——×ï/  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ^~Ì쇧 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1”Ô€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1PÐX Ø ` à h è $d €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll æ æ\\ÒÒ)H*H4¾4¾>2?2I¦I¦ST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllææ``ÚÚ T T Î ÎHHÂÂ<<¶¶,,¢¢ Ž Ž"#%z%z'ð'ð)f*f,Ü,Ü.R/R1È1È3>4>6´6´8*9*; ; =>@Š@ŠBþBþDrErGæGæIZJZLÎLÎNBOBQ¶Q¶S*T*VžVžXY[†[†]ú]ú_n`nbâbâdVeVgÊgÊi>j>l²l²n&o&qšqšstv‚v‚xöxözj{j}Þ}ÞR€R‚ƂƄ:…:‡®‡®‰!Š!Œ”Œ”Ž�‘z‘z“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1Å}!1ÐÅ}!1c¾!1Lc¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1‘ç!1(‘ç!1/("1¤/("1y/"1¬y/"19"1(9"19<"1ô9<"1cQ"1€cQ"1’"1ü’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1=Ú"1´ =Ú"1�þ"10 �þ"1ß+#1Pß+#1—:#1,—:#1³�#1˜³�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1«€&1Ì«€&1{R'1{R'1#e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘þÊ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @X˜ï/ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘þÊ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @X˜ï/ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘þÊ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @X˜ï/  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ççk‹  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1^ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1x ø n î d ä ZÚn® €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ú ÚOOÄÄ)9*94®4®>%?%IšIšST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÚÚOOÄÄ 9 9 ® ®##˜˜  ‚‚÷÷llááV V"Ë"Ë$@%@'µ'µ)***,Ÿ,Ÿ./1‰1‰3þ3þ5s6s8è8è:];]=Ò=Ò?G@GB¼B¼D1E1G¦G¦IJL’L’NOQ|Q|SñSñUfVfXÛXÛZP[P]Å]Å_:`:b¯b¯d$e$g™g™ijlƒlƒnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1~!1Ð~!1±¾!1L±¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1ßç!1(ßç!1}("1¤}("1y/"1¬y/"19"1(9"19<"1ô9<"1±Q"1€±Q"1O’"1üO’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1Ó:#1,Ó:#1-’#1˜-’#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1{€&1Ì{€&1KR'1KR'1ód'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™K b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 W——×ï/ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™K b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 W——×ï/ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™K  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 W——×ï/  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Uu³Óˆ¨ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1\œ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1p ð x ø €ˆì, €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âVVÊÊ)>*>4²4²>$?$I–I–ST^z^zhìhìr^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢ŠŠþþr r"æ"æ$Z%Z'Î'Î)B*B,¶,¶.*/*1ž1ž346†6†8ú8ú:n;n=â=â?T@TBÆBÆD8E8GªGªIJLŽLŽNOQrQrSäSäUVVVXÈXÈZ:[:]¬]¬_`b�b�degtgtiæiækXlXnÊnÊp‡>‰¯‰¯‹ Œ ޑޑ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1Ý}!1�Ý}!1{¾!1{¾!1¿Â!1$¿Â!1¯ç!1�¯ç!1M("1M("1‘,"1$‘,"1�Q"1��Q"1’"1’"1c–"1$c–"1Ù"1�Ù"1qý"1üqý"1;0#1 ;0#1#1| #1ó$1Ü ó$1Ëò$1  Ëò$1ß%1¤ ß%1)�&1„ )�&1ùR'1ÈùR'1¡e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations15o£ÞÍ™ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations15o£ÞÍ™ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations15o£ÞÍ™ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Çç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10zz ô ônnää)Z*Z4Ð4Ð>D?DI¸I¸S,T,^ ^ hisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðzzôônnèè b b Ü ÜVVÐÐJJÄÄ>>¸¸22 ¬ ¬"&#&%œ%œ'(*ˆ*ˆ,þ,þ.t/t1ê1ê3`4`6Ö6Ö8L9L;Â;Â=8>8@®@®B$C$EšEšGHJ„J„LøLøNlOlQàQàSTTTVÈVÈXI?IIÀIÀS5T5^¨^¨hisŒsŒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuîîggàà Y Y Ò ÒKKÄÄ==¶¶//¨¨!! š š"#%Œ%Œ'(*~*~,÷,÷.p/p1é1é3b4b6Û6Û8T9T;Í;Í=F>F@¿@¿B8C8E±E±G*H*J¡J¡LMO�O�QRT}T}VôVôXkYk[â[â]Y^Y`Ð`ÐbEcEeºeºg/h/j¤j¤lmoŒoŒqÿqÿsrtrvåvåxXyX{Ë{Ë}>~>€±€±‚$ƒ$…—…—‡ˆŠzŠzŒìŒìŽ^�^‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1Ý}!1�Ý}!1{¾!1{¾!1¿Â!1$¿Â!1©ç!1�©ç!1G("1G("1‘,"1$‘,"1{Q"1�{Q"1’"1’"1c–"1$c–"1Ù"1�Ù"1qý"1üqý"1;0#1 ;0#1#1| #1Û$1Ü Û$1³ò$1  ³ò$1íÞ%1¤ íÞ%1±€&1„ ±€&1�R'1È�R'1)e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations16m¡ÚМ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations16m¡ÚМ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations16m¡ÚМ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Çç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx ð ðhhÜÜ)P*P4Ä4Ä>6?6I¨I¨ST^Œ^Œhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðxxððhhàà X X Ð ÐHHÀÀ88°°((   � �"#%|%|'ð'ð)d*d,Ø,Ø.L/L1À1À34446¨6¨89;�;�=>@x@xBìBìD`E`GÔGÔIFJFL¸L¸N*O*QœQœSTV€V€XòXòZd[d]Ö]Ö_H`Hbºbºd,e,gžgžijl‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1 ~!1� ~!1«¾!1 «¾!1SÅ!1SÅ!1Ùç!1€Ùç!1w("1üw("1%/"1%/"1«Q"1p«Q"1I’"1ìI’"1÷˜"1ô÷˜"1ÍÚ"1`ÍÚ"1-ÿ"1Ì-ÿ"1;0#1ì;0#1#1L #1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1á€&1T á€&1±R'1˜±R'1Ye'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹=  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1–Ö€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹=  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1–Ö€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1W‹=  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1–Ö €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1‚¢  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu ì ìccÚÚ)Q*Q4È4È>=?=I²I²S%T%^–^–hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuììccÚÚ Q Q È È??¶¶--¤¤’’  € €"÷"÷$n%n'å'å)\*\,Ó,Ó.J/J1Á1Á38486¯6¯8&9&;�;�=>@‹@‹BCEyEyGðGðIeJeLÚLÚNOOOQÄQÄS9T9V®V®X#Y#[˜[˜] ^ `‚`‚bõbõdhehgÛgÛiNjNlÁlÁn2o2q£q£stv…v…xöxözg{g}Ø}ØI€I‚º‚º„+…+‡›‡›‰ Š Œ}Œ}ŽîŽî�_‘_“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1×}!1�×}!1u¾!1 u¾!1¹Â!1¹Â!1©ç!1€©ç!1G("1üG("1‹,"1‹,"1{Q"1p{Q"1’"1ì’"1]–"1ô]–"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"150#1ì50#1#1L #1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l Û b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l Û b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l Û b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1i‰  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss é é^^ÓÓ)H*H4»4»>.?.I¡I¡ST^‡^‡húhúrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðsséé__ÕÕ K K Á Á77­­##™™ƒƒøøm m"â"â$W%W'Ì'Ì)A*A,¶,¶.+/+1 1 346Š6Š8ÿ8ÿ:t;t=ç=ç?Z@ZBÍBÍD@E@G³G³I&J&L™L™N O QQSòSòUeVeXØXØZK[K]¾]¾_1`1b¤b¤degŠgŠiýiýkplpnãnãpVqVsÉsÉu???IµIµS+T+^¡^¡his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqééaaÙÙ Q Q É ÉAA¹¹11©©!!™™ ‰ ‰"#%y%y'ñ'ñ)i*i,á,á.Y/Y1Ñ1Ñ3I4I6Á6Á8999;±;±=)>)@¡@¡BCE‘E‘G H JJLõLõNkOkQáQáSWTWVÍVÍXCYC[¹[¹]/^/`¥`¥bce‘e‘ghj}j}lólónioiqßqßsUtUvËvËxAyA{·{·}-~-€£€£‚ƒ…�…�‡ˆŠ{Š{ŒñŒñŽg�g‘ݑݓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1×}!1�×}!1u¾!1 u¾!1¹Â!1¹Â!1£ç!1€£ç!1A("1üA("1‹,"1‹,"1uQ"1puQ"1’"1ì’"1]–"1ô]–"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"150#1ì50#1#1L #1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1E€&1T E€&1R'1˜R'1½d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jž × b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jž × b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jž × b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1jŠ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss é é\\ÏÏ)B*B4³4³>$?$I•I•ST^w^whèhèrWsW} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðsséé__ÕÕ K K Á Á77­­##™™  òòe e"Ø"Ø$K%K'¾'¾)1*1,¤,¤./1Š1Š3ý3ý5p6p8ã8ã:V;V=Ç=Ç?8@8B©B©DEG‹G‹IüIüKmLmNÞNÞPOQOSÀSÀU1V1X¢X¢Z[]„]„_õ_õafbfd×d×fHgHi¹i¹k*l*n›n›p q s}s}uîuîw_x_zÐzÐ|A}A²²�#‚#„”„”†‡‰s‰s‹ã‹ã�SŽS�Ã�Ã’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1 ~!1� ~!1«¾!1 «¾!1MÅ!1MÅ!1ßç!1€ßç!1}("1ü}("1/"1/"1±Q"1p±Q"1O’"1ìO’"1ñ˜"1ôñ˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"150#1ì50#1#1L #1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1ç€&1T ç€&1·R'1˜·R'1_e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”F b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1—×€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”F b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1—×€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”F b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1—× €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1�¡  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ç ç]]ÓÓ)I*I4¿4¿>3?3I§I§ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqçç]]ÓÓ I I ¿ ¿55««!!——  ƒƒùùo o"å"å$[%['Ñ'Ñ)G*G,½,½.3/31©1©346•6•8 9 ;�;�=÷=÷?m@mBãBãDYEYGÏGÏICJCL·L·N+O+QŸQŸSTV‡V‡XûXûZo[o]ã]ã_W`WbËbËd?e?g³g³i'j'l›l›noqƒqƒs÷s÷ukvkxßxßzS{S}Ç}Ç;€;‚¯‚¯„#…#‡—‡—‰ Š ŒŒŽóŽó�g‘g“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ñ}!1�Ñ}!1o¾!1 o¾!1¿Â!1¿Â!1£ç!1€£ç!1A("1üA("1‘,"1‘,"1uQ"1puQ"1’"1ì’"1c–"1ôc–"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1/0#1ì/0#1û~#1L û~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1ç€&1T ç€&1·R'1˜·R'1_e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i� Ø b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i� Ø b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i� Ø b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1f†  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ê ê__ÔÔ)I*I4¾4¾>2?2I¦I¦ST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrêêbbÚÚ R R Ê ÊBBºº22ªª””  ~ ~"ó"ó$h%h'Ý'Ý)R*R,Ç,Ç.e>g²g²i&j&lšlšnoq‚q‚sösöujvjxÞxÞzR{R}Æ}Æ:€:‚®‚®„"…"‡–‡–‰ Š Œ€Œ€ŽõŽõ�j‘j“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1±¾!1 ±¾!15Å!15Å!1åç!1€åç!1ƒ("1üƒ("1/"1/"1·Q"1p·Q"1U’"1ìU’"1Ù˜"1ôÙ˜"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1�&1T �&1çR'1˜çR'1�e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�M b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¿€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�M b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¿€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�M b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¿ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1k‹  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ä ä[[ÒÒ)I*I4À4À>5?5IªIªST^”^”h i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmää[[ÒÒ I I À À77®®%%œœŠŠ x x"ï"ï$f%f'Ý'Ý)T*T,Ë,Ë.B/B1¹1¹30406§6§89;•;•= > @ƒ@ƒBúBúDqEqGèGèI]J]LÒLÒNGOGQ¼Q¼S1T1V¦V¦XY[�[�]^`z`zbïbïddedgÙgÙiNjNlÃlÃn8o8q­q­s"t"v—v—x y {�{�}ö}ök€k‚à‚à„U…U‡ʇʉ?Š?Œ´Œ´Ž)�)‘ž‘ž“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ë}!1�Ë}!1i¾!1 i¾!1¿Â!1¿Â!1�ç!1€�ç!1;("1ü;("1‘,"1‘,"1oQ"1poQ"1 ’"1ì ’"1c–"1ôc–"1»Ú"1`»Ú"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›Ô b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›Ô b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›Ô b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1u•  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æ[[ÐÐ)E*E4º4º>-?-I I ST^„^„höhörhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  €€õõj j"ß"ß$T%T'É'É)>*>,³,³.(/(1�1�346‡6‡8ü8ü:q;q=æ=æ?[@[BÐBÐDEEEGºGºI-J-L L NOQ†Q†SùSùUlVlXßXßZR[R]Å]Å_8`8bªbªdegŽgŽijlrlrnänäpVqVsÈsÈu:v:x¬x¬z{}�}�€‚t‚t„æ„æ†X‡X‰ˉˋ>Œ>ޱޱ�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1±¾!1 ±¾!15Å!15Å!1åç!1€åç!1ƒ("1üƒ("1/"1/"1·Q"1p·Q"1U’"1ìU’"1Ù˜"1ôÙ˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1±€&1T ±€&1�R'1˜�R'1)e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�O b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¿€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�O b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¿€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�O b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¿ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1‚¢  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn ä äZZÏÏ)D*D4¹4¹>,?,IŸIŸST^…^…høhørksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnääZZÐÐ F F ¼ ¼22¨¨””  €€ööl l"â"â$W%W'Ì'Ì)A*A,¶,¶.+/+1 1 346Š6Š8ÿ8ÿ:t;t=é=é?^@^BÓBÓDHEHG½G½I0J0L£L£NOQ‰Q‰SüSüUoVoXâXâZU[U]È]È_;`;b®b®d!e!g”g”ijlzlzníníp`q`sÓsÓuFvFx¹x¹z,{,}Ÿ}Ÿ€‚…‚…„ø„ø†k‡k‰Þ‰Þ‹QŒQŽÄŽÄ�7‘7“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ë}!1�Ë}!1i¾!1 i¾!1³Â!1³Â!1�ç!1€�ç!1;("1ü;("1…,"1…,"1oQ"1poQ"1 ’"1ì ’"1W–"1ôW–"1�Ú"1`�Ú"1ýþ"1Ìýþ"1/0#1ì/0#1û~#1L û~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš Õ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš Õ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš Õ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1_  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn å å\\ÓÓ)J*J4Á4Á>8?8I¯I¯S&T&^˜^˜h i s|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnåå\\ÓÓ J J Á Á88¯¯&&��‹‹ y y"ð"ð$g%g'Þ'Þ)U*U,Ì,Ì.C/C1º1º31416¨6¨89;–;–= > @„@„BûBûDrErGéGéI`J`L×L×NNONQÅQÅS1?1I¥I¥ST^�^�hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmççaaÛÛ U U Ï ÏIIÃÃ==··--££ � �"#%{%{'ñ'ñ)g*g,Ý,Ý.S/S1É1É3?4?6µ6µ8+9+;¡;¡=>@‰@‰BýBýDqEqGåGåIYJYLÍLÍNAOAQµQµS)T)V�V�XY[…[…]ù]ù_m`mbábádUeUgÉgÉi=j=l±l±n%o%q™q™s t v�v�xõxõzi{i}Ý}ÝQ€Q‚łń9…9‡«‡«‰ŠŒ‘Œ‘Ž�‘w‘w“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Å}!1�Å}!1c¾!1 c¾!1³Â!1³Â!1—ç!1€—ç!15("1ü5("1…,"1…,"1iQ"1piQ"1’"1ì’"1W–"1ôW–"1»Ú"1`»Ú"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Ñ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Ñ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Ñ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1zš  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll á áVVËË)@*@4µ4µ>*?*IŸIŸST^†^†høhørjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllááVVËË @ @ µ µ**ŸŸ‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?N@NBÃBÃD8E8G­G­I"J"L—L—N O Q�Q�SöSöUkVkXàXàZU[U]Ê]Ê_?`?b´b´d)e)gžgžijlˆlˆnúnúplqlsÞsÞuPvPxÂxÂz4{4}¦}¦€‚ЂЄü„ü†n‡n‰à‰à‹RŒRŽÄŽÄ�6‘6“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1·¾!1 ·¾!1SÅ!1SÅ!1ëç!1€ëç!1‰("1ü‰("1%/"1%/"1½Q"1p½Q"1[’"1ì[’"1÷˜"1ô÷˜"1Ù"1`Ù"1eý"1Ìeý"1/0#1ì/0#1û~#1L û~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1{€&1T {€&1KR'1˜KR'1ód'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦X$ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¡á€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦X$ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¡á€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦X$ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¡á €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Çç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm å åYYÍÍ)A*A4³4³>%?%I—I—S T ^{^{híhír]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmåå]]ÕÕ M M Å Å==µµ--¥¥�� u u"é"é$]%]'Ñ'Ñ)E*E,¹,¹.-/-1¡1¡346‰6‰8ý8ý:q;q=ã=ã?U@UBÇBÇD9E9G«G«IJL�L�NOQsQsSåSåUWVWXÉXÉZ;[;]­]­_`b‘b‘deguguiçiçkYlYnËnËp=q=s¯s¯u!v!x“x“z{}w}wéé�[‚[„̈́͆=‡=‰®‰®‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Å}!1�Å}!1c¾!1 c¾!1¿Â!1¿Â!1‘ç!1€‘ç!1/("1ü/("1‘,"1‘,"1cQ"1pcQ"1’"1ì’"1c–"1ôc–"1—Ú"1`—Ú"1÷þ"1Ì÷þ"1)0#1ì)0#1õ~#1L õ~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ò b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ò b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ò b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1\|  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn è èbbØØ)N*N4Ä4Ä>8?8I¬I¬S T ^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnèèbbÜÜ V V Ð ÐJJÄÄ>>¸¸22¬¬&&    "#%�%�'(*|*|,ò,ò.h/h1Þ1Þ3T4T6Ê6Ê8@9@;¶;¶=,>,@¢@¢BCEŽEŽGHJxJxLìLìN`O`QÔQÔSHTHV¼V¼X0Y0[¤[¤]^`Œ`Œbcetetgègèi\j\lÐlÐnDoDq¸q¸s,t,v v xy{ˆ{ˆ}ü}üp€p‚ä‚ä„X…X‡̇̉@Š@Œ´Œ´Ž(�(‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1~!1�~!1½¾!1 ½¾!1/Å!1/Å!1ëç!1€ëç!1‰("1ü‰("1/"1/"1½Q"1p½Q"1[’"1ì[’"1Ó˜"1ôÓ˜"1Ù"1`Ù"1eý"1Ìeý"1)0#1ì)0#1õ~#1L õ~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%15�&1T 5�&1S'1˜S'1­e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯_+ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1€À€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯_+ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1€À€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯_+ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1€À €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Çç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii â â[[ÔÔ)M*M4Æ4Æ>=?=I´I´S)T)^œ^œhis€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiââ[[ÔÔ M M Æ Æ??¸¸11ªª##œœ Ž Ž"#%€%€'ù'ù)r*r,ë,ë.d/d1Ý1Ý3V4V6Ï6Ï8H9H;Á;Á=:>:@³@³B,C,E¥E¥GHJ•J•L M OƒOƒQúQúSqTqVèVèX_Y_[Ö[Ö]M^M`Ä`Äb9c9e®e®g#h#j˜j˜l m o€o€qóqósftfvÙvÙxLyL{¿{¿}2~2€¥€¥‚ƒ…‹…‹‡ü‡ü‰nŠnŒàŒàŽR�R‘đē b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1¿}!1�¿}!1]¾!1 ]¾!1¿Â!1¿Â!1‘ç!1€‘ç!1/("1ü/("1‘,"1‘,"1cQ"1pcQ"1’"1ì’"1c–"1ôc–"1µÚ"1`µÚ"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Î b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Î b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Î b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1y™  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ä ä\\ÐÐ)D*D4¸4¸>*?*IœIœST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllää\\ÔÔ L L Ä Ä<<´´,,¤¤””  „ „"ü"ü$p%p'ä'ä)X*X,Ì,Ì.@/@1´1´3(4(6œ6œ89;„;„=ø=ø?l@lBàBàDTETGÈGÈI:J:L¬L¬NOQ�Q�STVtVtXæXæZX[X]Ê]Ê_<`v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1~!1�~!1½¾!1 ½¾!1/Å!1/Å!1ñç!1€ñç!1�("1ü�("1/"1/"1ÃQ"1pÃQ"1a’"1ìa’"1Ó˜"1ôÓ˜"1Ù"1`Ù"1eý"1Ìeý"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯a- b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1€À€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯a- b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1€À€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯a- b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1€À €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Çç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii à àWWÎÎ)E*E4¼4¼>1?1I¦I¦ST^Š^Šhûhûrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiààWWÎÎ E E ¼ ¼33ªª!!˜˜††ýýt t"ë"ë$b%b'Ù'Ù)P*P,Ç,Ç.>/>1µ1µ3,4,6£6£89;‘;‘=>@@BöBöDmEmGäGäIYJYLÎLÎNCOCQ¸Q¸S-T-V¢V¢XY[Œ[Œ]^`v`vbébéd\e\gÏgÏiBjBlµlµn&o&q—q—stvyvyxêxêz[{[}Ì}Ì=€=‚®‚®„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¿}!1�¿}!1]¾!1 ]¾!1§Â!1§Â!1‹ç!1€‹ç!1)("1ü)("1y,"1y,"1]Q"1p]Q"1û‘"1ìû‘"1K–"1ôK–"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ï b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ï b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ï b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Çç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝRRÇÇ)<*<4¯4¯>"?"I•I•ST^{^{hîhîr_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��wwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Û=Û?N@NBÁBÁD4E4G§G§IJL�L�NOQsQsSæSæUYVYXÌXÌZ?[?]²]²_%`%b˜b˜d e g~g~iñiñkdldn×n×pJqJs½s½u0v0x£x£z{}‰}‰üü�o‚o„â„â†S‡S‰ʼnŋ7Œ7ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1~!1�~!1½¾!1 ½¾!1GÅ!1GÅ!1ñç!1€ñç!1�("1ü�("1/"1/"1ÃQ"1pÃQ"1a’"1ìa’"1ë˜"1ôë˜"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1#0#1ì#0#1ï~#1L ï~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1A�&1T A�&1S'1˜S'1¹e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸h4 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1œÜ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸h4 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1œÜ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1„¸h4 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1œÜ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Çç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝUUÍÍ)E*E4½4½>3?3I©I©ST^•^•h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%���� } }"õ"õ$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=>@•@•B C E…E…GýGýIsJsLéLéN_O_QÕQÕSKTKVÁVÁX7Y7[­[­]#^#`™`™bce…e…gûgûiqjqlçlçn]o]qÓqÓsItIv¿v¿x5y5{«{«}!~!€—€—‚ ƒ …ƒ…ƒ‡ù‡ù‰oŠoŒåŒåŽ[�[‘ёѓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¹}!1�¹}!1W¾!1 W¾!1§Â!1§Â!1‹ç!1€‹ç!1)("1ü)("1y,"1y,"1]Q"1p]Q"1û‘"1ìû‘"1K–"1ôK–"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1#0#1ì#0#1ï~#1L ï~#1¥$1¬ ¥$1}ò$1p }ò$1·Þ%1t ·Þ%1'€&1T '€&1÷Q'1˜÷Q'1Ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ÿË b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ÿË b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’ÿË b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Çç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝPPÃÃ)6*64§4§>?I‰I‰SúSú]k^khÜhÜrKsK} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��ssææY Y"Ì"Ì$?%?'²'²)%*%,˜,˜. / 1~1~3ñ3ñ5d6d8×8×:J;J=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éaZbZdËdËf'?'I›I›ST^ƒ^ƒh÷h÷rksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡‹‡‹‰ÿ‰ÿ‹sŒsŽçŽç�[‘[“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1³}!1�³}!1Q¾!1 Q¾!1³Â!1³Â!1…ç!1€…ç!1#("1ü#("1…,"1…,"1WQ"1pWQ"1õ‘"1ìõ‘"1W–"1ôW–"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ì b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ì b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ì b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Çç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞSSÈÈ)=*=4²4²>&?&IšIšST^„^„høhørlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžˆˆýýr r"ç"ç$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`JbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡Œ‡Œ‰ŠŒvŒvŽëŽë�`‘`“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1%~!1�%~!1þ!1 þ!1#Å!1#Å!1÷ç!1€÷ç!1•("1ü•("1õ."1õ."1ÉQ"1pÉQ"1g’"1ìg’"1ǘ"1ôǘ"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Áq= b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¿€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Áq= b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¿€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Áq= b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¿ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Çç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>)?)IžIžST^’^’his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜIQJQLÆLÆN;O;Q°Q°S%T%VšVšXY[„[„]ù]ù_n`nbãbãdXeXgÍgÍiBjBl·l·n6o6q«q«s t v•v•x y {{}ô}ôi€i‚Þ‚Þ„S…S‡ȇȉ=Š=Œ²Œ²Ž'�'‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1³}!1�³}!1Q¾!1 Q¾!1³Â!1³Â!1ç!1€ç!1("1ü("1…,"1…,"1QQ"1pQQ"1ï‘"1ìï‘"1W–"1ôW–"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�üÈ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�üÈ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�üÈ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Çç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94®4®>!?!I”I”S T ^|^|hîhîr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?O@OBÄBÄD9E9G®G®I!J!L”L”NOQzQzSíSíU`V`XÓXÓZF[F]¹]¹_,`,b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†P‡P‰ÉË6Œ6ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1+~!1�+~!1ɾ!1 ɾ!1#Å!1#Å!1÷ç!1€÷ç!1•("1ü•("1õ."1õ."1ÉQ"1pÉQ"1g’"1ìg’"1ǘ"1ôǘ"1ùØ"1`ùØ"1Yý"1ÌYý"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ás? b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¿€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ás? b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¿€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ás? b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¿ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Çç  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÃÃ)8*84­4­> ? I“I“ST^�^�hôhôrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD.?.I¥I¥ST^š^šh i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLÍLÍNDODQ»Q»S2T2V©V©X Y [—[—]^`…`…bübüdsesgêgêiajalØlØnVoVqÈqÈs:t:v¬v¬xy{�{�}~€t€t‚æ‚æ„X…X‡ʇʉ<Š<Œ®Œ®Ž � ‘’‘’“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!1+~!1�+~!1ɾ!1 ɾ!1;Å!1;Å!1ýç!1€ýç!1›("1ü›("1 /"1 /"1ÏQ"1pÏQ"1m’"1ìm’"1ߘ"1ôߘ"1ÕØ"1`ÕØ"15ý"1Ì5ý"10#1ì0#1é~#1L é~#1Ï$1¬ Ï$1÷ð$1p ÷ð$1ÕÀ%1t ÕÀ%1ç€&1T ç€&1ÝP'1˜ÝP'1…c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–ÊzF b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1šÚ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–ÊzF b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1šÚ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–ÊzF b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1šÚ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Çà  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Û ÛQQÑÑ)G*G4»4»>/?/I£I£S#T#^—^—h i s}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÛÛUUÏÏ I I à Ã==··11««!!——  ƒ ƒ"ù"ù$y%y'ï'ï)e*e,Û,Û.Q/Q1Ç1Ç3=4=6³6³8)9);Ÿ;Ÿ=>@‡@‡BûBûDoEoGãGãIWJWLËLËN?O?Q³Q³S'T'V›V›XY[ƒ[ƒ]÷]÷_k`kbëbëd_e_gÓgÓiGjGl»l»n/o/q£q£stv‹v‹xÿxÿzs{s}ç}ç[€[‚ςτC…C‡µ‡µ‰(Š(Œ›Œ›Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!1­}!1�­}!1K¾!1 K¾!1¡Â!1¡Â!1yç!1€yç!1("1ü("1s,"1s,"1KQ"1pKQ"1é‘"1ìé‘"1E–"1ôE–"1OÚ"1`OÚ"1¯þ"1Ô¯þ"10#1ô0#1é~#1T é~#1½$1´ ½$1åð$1x åð$1ÃÀ%1| ÃÀ%1Ÿ€&1\ Ÿ€&1•P'1 •P'1=c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒùÅ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒùÅ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1XŒùÅ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1êzÑ=  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Õ ÕJJÃÃ)8*84­4­>"?"I—I—S T ^Š^Šhühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÕÕJJ¿¿ 4 4 © ©““}}òòggÜÜQ Q"Æ"Æ$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nþnþppqpsâsâuTvTxÆxÆz8{8}ª}ª€‚ނބ…‡r‡r‰ä‰ä‹VŒVŽÈŽÈ�:‘:“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!1+~!1�+~!1ɾ!1 ɾ!1;Å!1;Å!1ýç!1€ýç!1›("1ü›("1 /"1 /"1ÏQ"1pÏQ"1m’"1ìm’"1ߘ"1ôߘ"1ÕØ"1`ÕØ"15ý"1Ì5ý"10#1ì0#1é~#1L é~#1·$1¬ ·$1ßð$1p ßð$1½À%1t ½À%1o€&1T o€&1eP'1˜eP'1 c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–Ê|H b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1›Û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–Ê|H b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1›Û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1–Ê|H b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1›Û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Çà  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ù ÙMMÉÉ)=*=4¯4¯>!?!I“I“ST^ƒ^ƒhõhõrese} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A ¹ ¹11©©!!™™  ��õõi i"Ý"Ý$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b™b™d e g}g}iïiïkalanÓnÓpEqEs·s·u)v)x›x›z { }}ññ�c‚c„Õ„Õ†E‡E‰¶‰¶‹'Œ'Ž˜Ž˜� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„1i!1e}!1e}!1¾!1|¾!1gÃ!1„gÃ!1�Æ!1ð�Æ!1'Ð!1|'Ð!1]Ó!1H]Ó!17ç!1Ô7ç!1Õ'"1PÕ'"19-"1X9-"1o0"1Äo0"1ù9"1Pù9"1/="1/="1 Q"1¨ Q"1§‘"1$ §‘"1 —"1,  —"1Aš"1˜ Aš"1Ë£"1$ Ë£"1§"1ð §"1CÔ"1| CÔ"1IÚ"1XIÚ"1©þ"1Ô©þ"1ë+#1ôë+#1…:#1Ð…:#1Ao#1\Ao#1ߎ#1€ߎ#1Ý™#1¬Ý™#1Û$18Û$1³ò$1ü³ò$1íÞ%1íÞ%1gì%1àgì%1A�&1lA�&1S'1°S'1¹e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PÁ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1É  bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PÁ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1É   bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PÁ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1É  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 É銪$D bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1ZY™ò2‹Ë$d  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dšYÙ2²õu ‹ÐPäd«+½=–oïHÈ!¡úzÓS ¬ , …  ^ Þ 7 ·  � é )Ê ‚Â[´ô bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ × ×ÑÑ)I*I4Á4Á>9?9I¯I¯S%T%^§^§his“s“} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__××OOÇÇ ? ? · ·//§§——ÓÓKKÃÃ; ;"³"³$q%q'é'é)a*a,Ù,Ù.Q/Q1É1É3A4A6¹6¹8191;©;©=!>!@™@™BCE‰E‰GHJyJyLïLïNeOeQÛQÛSQTQVÇVÇX=Y=[³[³])^)`Ÿ`Ÿbce‹e‹ghjwjwlílínoooqåqås[t[vÑvÑxGyG{½{½}3~3€©€©‚ƒ…•…•‡ ˆ Š�Š�Œ÷Œ÷Žm�m‘ã‘ã“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1C~!1èC~!1á¾!1dá¾!1�Å!1l�Å!1Ï!1èÏ!1OÒ!1´OÒ!1è!1@è!1³("1¼³("1a/"1Äa/"1ë8"1@ë8"1!<"1 !<"1çQ"1˜çQ"1…’"1…’"13™"1 3™"1½¢"1˜ ½¢"1ó¥"1d ó¥"15Ó"1ð 5Ó"1£Ú"1Ì £Ú"1ÿ"1H ÿ"1E,#1hE,#1…:#1D…:#1Ao#1ÐAo#1ߎ#1ôߎ#1Ý™#1 Ý™#1É$1¬É$1¡ò$1p¡ò$1ÛÞ%1tÛÞ%1Uì%1TUì%1ó€&1àó€&1ÃR'1$ÃR'1ke'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1²æ˜d b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™m­ò2 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1²æ˜d b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™m­ò2 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1²æ˜d  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™m­ò2  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 |œ(}� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1VY™ò2‹Ë$d  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d–YÙëk2²À@ ‹•äd½=–oïHÈ!¡úzÓS ¬ , …  ^ Þ 7 ·  � f ¦ é )‚Â[´ô bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPP��)@*@4·4·>0?0I§I§ST^œ^œhis€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$!%!'˜'˜)*,†,†.ý.ý0°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIXJXLÏLÏNFOFQ½Q½S4T4V«V«X"Y"[™[™]^`‡`‡bþbþdueugìgìicjclÚlÚnXoXqÊqÊsŠ>Œ°Œ°Ž"�"‘”‘”“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„1i!1e}!1e}!1¾!1|¾!1gÃ!1„gÃ!1�Æ!1ð�Æ!1'Ð!1|'Ð!1]Ó!1H]Ó!11ç!1Ô1ç!1Ï'"1PÏ'"19-"1X9-"1o0"1Äo0"1ù9"1Pù9"1/="1/="1Q"1¨Q"1¡‘"1$ ¡‘"1 —"1,  —"1Aš"1˜ Aš"1Ë£"1$ Ë£"1§"1ð §"1CÔ"1| CÔ"1Ú"1XÚ"1sþ"1Ôsþ"1µ+#1ôµ+#1…:#1Ð…:#1Ao#1\Ao#1ߎ#1€ߎ#1Ý™#1¬Ý™#1Ã$18Ã$1›ò$1ü›ò$1ÕÞ%1ÕÞ%1Oì%1àOì%1É€&1lÉ€&1™R'1°™R'1Ae'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¾Š b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1É  bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¾Š b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1É   bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1O¾Š  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1É  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÉéjŠí  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Y™ò2µõ‹Ë$d  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dYÙ2² ‹äd½=†–aáoï<¼HÈ—!¡úzÓS ¬ , …  ^ Þ 7 ·  � é )‚Âe¥[´ô bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Õ ÕÉÉ)?*?4µ4µ>+?+IŸIŸST^“^“his{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……¾¾44ªª "–"–$Q%Q'Ç'Ç)=*=,³,³.)/)1Ÿ1Ÿ346‹6‹89;w;w=í=í?c@cBÙBÙDOEOGÅGÅI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÃbÃd7e7g«g«ijl“l“noq‡q‡sûsûuovoxãxãzW{W}Ë}Ë?€?‚³‚³„'…'‡›‡›‰ŠŒƒŒƒŽ÷Ž÷�k‘k“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1I~!1èI~!1ç¾!1dç¾!1�Å!1l�Å!1Ï!1èÏ!1OÒ!1´OÒ!1è!1@è!1³("1¼³("1a/"1Äa/"1ë8"1@ë8"1!<"1 !<"1çQ"1˜çQ"1…’"1…’"13™"1 3™"1½¢"1˜ ½¢"1ó¥"1d ó¥"15Ó"1ð 5Ó"1sÚ"1Ì sÚ"1Óþ"1H Óþ"1,#1h,#1…:#1D…:#1Ao#1ÐAo#1ëŽ#1ôëŽ#1é™#1 é™#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1Ÿ€&1àŸ€&1oR'1$oR'1e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1´èœh b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™n®ò2 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1´èœh b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™n®ò2 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1´èœh  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™n®ò2  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w—Ll> bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Zš«ëô4�Í&f  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dZÚ4´ŽèhjêÄD?¿ž”xøéiRÒ+«„Ý] ¶ 6 �  h è A Á  š ô 4û ;ŽÎ'gÀ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Õ ÕJJ††)8*84­4­>$?$I™I™ST^Œ^Œhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÕÕJJ¿¿ 4 4 © ©““}}òòggÜÜQ Q"Æ"Æ$%'w'w)ì)ì+a,a.Ö.Ö0ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚noqrqrsäsäuVvVxÈxÈz:{:}¬}¬€‚�‚�„…‡t‡t‰æ‰æ‹XŒXŽÊŽÊ�<‘<“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1­}!1è­}!1K¾!1dK¾!1•Å!1l•Å!1Ï!1èÏ!1UÒ!1´UÒ!1ç!1@ç!1("1¼("1g/"1Äg/"1ñ8"1@ñ8"1'<"1 '<"1QQ"1˜QQ"1ï‘"1ï‘"19™"1 9™"1â"1˜ â"1ù¥"1d ù¥"1;Ó"1ð ;Ó"1OÚ"1Ì OÚ"1¯þ"1H ¯þ"1ñ+#1hñ+#1‹:#1D‹:#1Go#1ÐGo#1åŽ#1ôåŽ#1ã™#1 ã™#1É$1¬É$1¡ò$1p¡ò$1ÛÞ%1tÛÞ%1Uì%1TUì%1ÿ€&1àÿ€&1ÏR'1$ÏR'1we'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�þÊ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Zšó3 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�þÊ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Zšó3 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�þÊ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Zšó3  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Éé‹«&F bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Y™Yò2‹Ë$d  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥z©*©b­reasonOperatorInputsvalue8kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(TransposedRight), tiling: TensorTiling { tile_count: 192, replicas: 1, axes: [AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(2), partitions: 12, block_size: 64, padding_multiple: 64, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements4608input0Elements4608invocations1€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dYÙ™2²òr ‹ËKäd¤$½=–oïHÈ!¡úzÓS ¬ , …  ^ Þ 7 ·  � é )© é ‚Â[´ô bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Ö ÖMMÄÄ);*;4²4²>)?)IžIžS×S×]’^’his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÖÖMMÄÄ ; ; ² ²))  ŽŽ||óój j"á"á$X%X'Ï'Ï)F*F,½,½.4/41«1«3"4"6™6™89;‡;‡=þ=þ?u@uBìBìDcEcGÚGÚIQJQLÆLÆN;O;Q°Q°S%T%VšVšXY[„[„]ù]ù_n`nb§b§deg‘g‘ijl{l{n6o6q«q«s t v•v•x y {{}ô}ôi€i‚Þ‚Þ„S…S‡ȇȉ=Š=Œ²Œ²Ž'�'‘œ‘œ“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1=~!1è=~!1Û¾!1dÛ¾!1•Å!1l•Å!1Ï!1èÏ!1UÒ!1´UÒ!1 è!1@ è!1§("1¼§("1g/"1Äg/"1ñ8"1@ñ8"1'<"1 '<"1ÛQ"1˜ÛQ"1y’"1y’"19™"1 9™"1â"1˜ â"1ù¥"1d ù¥"1;Ó"1ð ;Ó"1ÇÚ"1Ì ÇÚ"1'ÿ"1H 'ÿ"1i,#1hi,#1‹:#1D‹:#1Go#1ÐGo#1åŽ#1ôåŽ#1ã™#1 ã™#1Ã$1¬Ã$1›ò$1p›ò$1ÕÞ%1tÕÞ%1Oì%1TOì%1ÿ€&1àÿ€&1ÏR'1$ÏR'1we'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1§Û�Y b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™«ëò2 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1§Û�Y b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™«ëò2 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1§Û�Y  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™«ëò2  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 {›&˜¸ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Y™ Mò2‹Ë$d  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dYÙ �äd2²»; ‹’äd½=–oïHÈ!¡úzÓS ¬ , …  ^ Þ 7 ·  � } ½ é )‚Â[´ô bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff à àZZÐÐ)F*F4¼4¼>2?2I¦I¦ST^T^Thisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffààZZÔÔ N N È ÈBB¼¼66°°**¤¤ ˜ ˜"#%ˆ%ˆ'þ'þ)t*t,ê,ê.`/`1Ö1Ö3L4L6Â6Â8898;®;®=$>$@š@šBCE†E†GüGüIrJrLæLæNZOZQÎQÎSBTBV¶V¶X*Y*[ž[ž]^`†`†búbúdnengâgâiVjVlÊlÊnoqxqxsìsìu`v`xÔxÔz„{„}ø}øl€l‚à‚à„T…T‡ȇȉ<Š<Œ°Œ°Ž$�$‘˜‘˜“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1­}!1è­}!1K¾!1dK¾!1•Å!1l•Å!1Ï!1èÏ!1UÒ!1´UÒ!1yç!1@yç!1("1¼("1g/"1Äg/"1ñ8"1@ñ8"1'<"1 '<"1KQ"1˜KQ"1é‘"1é‘"19™"1 9™"1â"1˜ â"1ù¥"1d ù¥"1;Ó"1ð ;Ó"1Ú"1Ì Ú"1yþ"1H yþ"1»+#1h»+#1‹:#1D‹:#1Go#1ÐGo#1åŽ#1ôåŽ#1ã™#1 ã™#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1™€&1à™€&1iR'1$iR'1e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�üÈ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Zšó3 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�üÈ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Zšó3 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�üÈ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Zšó3  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Éék‹ó bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Y™ò2²ò‹Ë$d  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dYÙ2² ‹äd½=}ý–VÖoï/¯HȈ!¡úzÕU ® . ‡  ` à 9 ¹  ’ ë +B‚„Ä]¶ö bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ö ÖLLÁÁ)6*64«4«> ? I“I“SÉSÉ]�^�hôhôrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††üürrèè^ ^"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I$J$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/bebedØdØfKgKi¾i¾k1l1nénép\q\sÏsÏuBvBxµxµz({(}›}›€‚�‚�„ô„ô†g‡g‰Ú‰Ú‹MŒMŽÀŽÀ�3‘3“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1=~!1è=~!1Û¾!1dÛ¾!1•Å!1l•Å!1Ï!1èÏ!1UÒ!1´UÒ!1è!1@è!1­("1¼­("1g/"1Äg/"1ñ8"1@ñ8"1'<"1 '<"1áQ"1˜áQ"1’"1’"19™"1 9™"1â"1˜ â"1ù¥"1d ù¥"1;Ó"1ð ;Ó"1�Ú"1Ì �Ú"1ýþ"1H ýþ"1?,#1h?,#1‹:#1D‹:#1Go#1ÐGo#1åŽ#1ôåŽ#1ã™#1 ã™#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1‡€&1à‡€&1WR'1$WR'1ÿd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1©Ý‘] b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™ªêò2 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1©Ý‘] b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™ªêò2 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1©Ý‘]  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™ªêò2  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 zš$Dq‘ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1Y™¤äò2‹Ë$d  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dYÙ2² ‹ädié½=@À–—qñînJÊ#c¼üUÕ. ®  ‡ à ` ¹ 9 ’  k ë DÄT]¶ö bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜTTÈÈ)<*<4°4°>$?$I–I–ST^A^Ahðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœŒŒ | |"ô"ô$h%h'Ü'Ü)P*P,Ä,Ä.8/81¬1¬3 4 6”6”89;|;|=ð=ð?d@dBØBØDLELGÀGÀI4J4L¦L¦NOQŠQŠSüSüUnVnXàXàZR[R]Ä]Ä_6`6b¨b¨degŒgŒiþiþkplpn©n©pqs�s�uÿuÿwqxqz { }’}’€‚v‚v„è„è†Z‡Z‰̉̋>Œ>ްް�"‘"“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1³}!1г}!1Q¾!1LQ¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1ç!1(ç!1("1¤("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1QQ"1€QQ"1ï‘"1üï‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#1{2#1,{2#1·Œ#1˜·Œ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Û€&1ÌÛ€&1«R'1«R'1Se'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽÿË b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 UY™ò2 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽÿË b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 UY™ò2 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽÿË  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 UY™ò2  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Œ¬Zz(H bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1V€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1–ímÄD›† Æ  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ý ÝSSÉÉ)?*?4µ4µ>)?)I�I�ST^…^…hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÝÝWWÑÑ K K Å Å??¹¹33­­##™™ … …"û"û$q%q'ç'ç)]*],Ó,Ó.I/I1¿1¿35456«6«8!9!;—;—= > @�@�BõBõDiEiGÝGÝIQJQLÅLÅN9O9Q­Q­S!T!V•V•X Y [}[}]ñ]ñ_e`ebÙbÙdMeMgÁgÁi5j5l©l©noq‘q‘stvyvyxíxíza{a}Õ}ÕI€I‚½‚½„1…1‡¥‡¥‰ŠŒ‹Œ‹ŽþŽþ�q‘q“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!17~!1Ð7~!1Õ¾!1LÕ¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1è!1(è!1¡("1¤¡("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1ÕQ"1€ÕQ"1s’"1üs’"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"11Ú"1´ 1Ú"1‘þ"10 ‘þ"1Ó+#1PÓ+#1i2#1,i2#1ÃŒ#1˜ÃŒ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐ‚N b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™}½ö6 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐ‚N b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™}½ö6 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐ‚N  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™}½ö6  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 v–Éé# bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1„Ý]¶6�” Ô  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84­4­> ? I“I“ST^y^yhìhìr_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?L@LB¿B¿D2E2G¥G¥IJL‹L‹NþNþPqQqSäSäUWVWXÊXÊZ=[=]°]°_#`#b–b–d e g|g|iïiïkblbnÕnÕpHqHs»s»u.v.x¡x¡z{}‡}‡úú�m‚m„à„à†S‡S‰ʼnŋ7Œ7ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1­}!1Э}!1K¾!1LK¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1ç!1(ç!1("1¤("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1QQ"1€QQ"1ï‘"1üï‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1gÚ"1´ gÚ"1Çþ"10 Çþ"1 ,#1P ,#15#1,5#1M�#1˜M�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1c€&1Ìc€&13R'13R'1Ûd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽýÉ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 VY™ò2 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽýÉ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 VY™ò2 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽýÉ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 VY™ò2  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 lŒ5Uù bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1­í€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ròIÉ  ÷w] €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛOOÃÃ)7*74«4«>?I�I�ST^s^shåhårWsW} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››ƒƒ÷÷k k"ß"ß$S%S'Ç'Ç);*;,¯,¯.#/#1—1—3 4 668ó8ó:g;g=Û=Û?M@MB¿B¿D1E1G£G£IJL‡L‡NùNùPkQkSÝSÝUOVOXÁXÁZ3[3]¥]¥_`b‰b‰dûdûfmgmißißkQlQnÃnÃp5q5s§s§uvx‹x‹zýzý|o}oáá�S‚S„ńņ7‡7‰¨‰¨‹ŒŽŠŽŠ�û�û’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!17~!1Ð7~!1Õ¾!1LÕ¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1 è!1( è!1§("1¤§("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1ÛQ"1€ÛQ"1y’"1üy’"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1UÚ"1´ UÚ"1µþ"10 µþ"1÷+#1P÷+#15#1,5#1_�#1˜_�#1Ÿ$1$Ÿ$1wò$1èwò$1±Þ%1ì±Þ%1!€&1Ì!€&1ñQ'1ñQ'1™d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒ†R b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™}½÷7 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒ†R b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™}½÷7 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒ†R  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™}½÷7  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 }�Éé*J bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Ý€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hèAÁšós-m €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙLL¿¿)2*24¥4¥>?I‡I‡SøSø]i^ihÚhÚrKsK} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰üüooââU U"È"È$;%;'®'®)!*!,”,”./1z1z3í3í5`6`8Ó8Ó:F;F=¹=¹?*@*B›B›D E G}G}IîIîK_L_NÐNÐPAQAS²S²U#V#X”X”Z[]v]v_ç_çaXbXdÉdÉf:g:i«i«kln�n�pþpþrosouàuàwQxQzÂzÂ|3}3¤¤�‚„†„††÷†÷ˆg‰g‹׋×�GŽG�·�·’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¹}!1й}!1W¾!1LW¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1…ç!1(…ç!1#("1¤#("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1WQ"1€WQ"1õ‘"1üõ‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1¯Ú"1´ ¯Ú"1ÿ"10 ÿ"1Q,#1PQ,#1o2#1,o2#1�Œ#1˜�Œ#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1/�&1Ì/�&1ÿR'1ÿR'1§e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�Ì b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @[›ô4 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�Ì b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @[›ô4 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�Ì  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @[›ô4  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 �­*Jy™ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 I€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ‰Þ^³3ˆY ™  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ü ÜUUÎÎ)G*G4À4À>9?9I°I°S%T%^˜^˜h i s~s~} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÜÜUUÎÎ G G À À99²²++¤¤–– ˆ ˆ"#%z%z'ó'ó)l*l,å,å.^/^1×1×3P4P6É6É8B9B;»;»=4>4@­@­B&C&EŸEŸGHJ‘J‘LMOOQöQöSmTmVäVäX[Y[[Ò[Ò]I^I`À`Àb5c5eªeªghj”j”l m o|o|qïqïsbtbvÕvÕxHyH{»{»}.~.€¡€¡‚ƒ…‡…‡‡ú‡ú‰lŠlŒތގP�P‘‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1+~!1Ð+~!1ɾ!1Lɾ!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1ýç!1(ýç!1›("1¤›("1/"1¬/"1 9"1( 9"1?<"1ô?<"1ÏQ"1€ÏQ"1m’"1üm’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1Ú"1´ Ú"1þ"10 þ"1Á+#1PÁ+#1i2#1,i2#1#�#1˜#�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘ÅwC b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™ àA bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘ÅwC b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™ àA bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘ÅwC  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™ àA  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 i‰Ééñ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€à`À@  ð 0 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞSSÈÈ)=*=4²4²>(?(IœIœST^„^„høhørnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžˆˆýýr r"ç"ç$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡އމŠŒxŒxŽíŽí�b‘b“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1³}!1г}!1Q¾!1LQ¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1…ç!1(…ç!1#("1¤#("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1WQ"1€WQ"1õ‘"1üõ‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1‹Ú"1´ ‹Ú"1ëþ"10 ëþ"1-,#1P-,#15#1,5#1#�#1˜#�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�þÊ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Zšó3 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�þÊ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Zšó3 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�þÊ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Zšó3  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 m�ýXx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1žÞ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1]Ý2²‡Ü\î . €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>-?-I¢I¢ST^†^†h÷h÷risi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbåbådXeXgËgËi>j>l±l±n"o"q“q“stvuvuxæxæzW{W}È}È9€9‚ª‚ª„…‡�‡�‰þ‰þ‹oŒoŽàŽà�Q‘Q“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!11~!1Ð1~!1Ͼ!1LϾ!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1è!1(è!1¡("1¤¡("1/"1¬/"1 9"1( 9"1?<"1ô?<"1ÕQ"1€ÕQ"1s’"1üs’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#1 5#1, 5#1Å�#1˜Å�#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Ç{G b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™ŸßB bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Ç{G b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™ŸßB bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Ç{G  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™ŸßB  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x˜Éé @ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 à€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€`à@À  �Ð €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94®4®>#?#I–I–ST^z^zhìhìr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?O@OBÄBÄD9E9G®G®I#J#L–L–N O Q|Q|SïSïUbVbXÕXÕZH[H]»]»_.`.b b deg„g„iöiökhlhnÚnÚpLqLs¾s¾u0v0x¢x¢z{}†}†øø�j‚j„܄܆P‡P‰ÉË6Œ6ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¹}!1й}!1W¾!1LW¾!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1‹ç!1(‹ç!1)("1¤)("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1]Q"1€]Q"1û‘"1üû‘"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1µÚ"1´ µÚ"1ÿ"10 ÿ"1W,#1PW,#1i2#1,i2#1·Œ#1˜·Œ#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1A�&1ÌA�&1S'1S'1¹e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�Í b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Iq± J bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�Í b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Iq± J bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�Í  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Iq± J  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 „¤8|œ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€×W®.…€ À  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSËË)C*C4»4»>3?3I©I©ST^•^•h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$k%k'ã'ã)[*[,Ó,Ó.K/K1Ã1Ã3;4;6³6³8+9+;£;£=>@“@“B C EƒEƒGûGûIsJsLéLéN_O_QÕQÕSKTKVÁVÁX7Y7[­[­]#^#`™`™bce…e…gûgûiqjqlçlçn]o]qÓqÓsItIv¿v¿x5y5{«{«}!~!€—€—‚ ƒ …ƒ…ƒ‡ù‡ù‰oŠoŒåŒåŽ[�[‘ёѓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1%~!1Ð%~!1þ!1Lþ!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1÷ç!1(÷ç!1•("1¤•("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1ÉQ"1€ÉQ"1g’"1üg’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1 Ú"1´  Ú"1mþ"10 mþ"1¯+#1P¯+#1i2#1,i2#1×�#1˜×�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1ù€&1Ìù€&1ÉR'1ÉR'1qe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºl8 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™}½õ5 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºl8 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™}½õ5 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºl8  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™}½õ5  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Éé\|Ûû bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ëkÖVÁA à €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ý ÝTTËË)B*B4¹4¹>2?2I©I©S T ^’^’hisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÝÝTTËË B B ¹ ¹00§§••  ƒƒúúq q"è"è$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIZJZLÑLÑNHOHQ¿Q¿S6T6V­V­X$Y$[›[›]^`‰`‰bcewewgîgîiejelÜlÜnNoNqÀqÀs2t2v¤v¤xy{ˆ{ˆ}ú}úl€l‚Þ‚Þ„P…P‡‡‰4Š4Œ¦Œ¦Ž�‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¹}!1й}!1W¾!1LW¾!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1‹ç!1(‹ç!1)("1¤)("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1]Q"1€]Q"1û‘"1üû‘"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#15#1,5#1M�#1˜M�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ÿË b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jp° I bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ÿË b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jp° I bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�ÿË  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jp° I  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 d„í m� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1—×€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1\Ü3³ ŠáaW €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙOOÅÅ);*;4±4±>'?'I›I›ST^ƒ^ƒh÷h÷rksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰ÿÿuuëëa a"×"×$M%M'Ã'Ã)9*9,¯,¯.%/%1›1›346‡6‡8ý8ý:s;s=é=é?_@_BÕBÕDKEKGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡‹‡‹‰ÿ‰ÿ‹sŒsŽçŽç�[‘[“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1+~!1Ð+~!1ɾ!1Lɾ!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1ýç!1(ýç!1›("1¤›("1y/"1¬y/"19"1(9"19<"1ô9<"1ÏQ"1€ÏQ"1m’"1üm’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1CÚ"1´ CÚ"1£þ"10 £þ"1å+#1På+#15#1,5#1…Ž#1˜…Ž#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Ÿ€&1ÌŸ€&1oR'1oR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼p< b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™}½ö6 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼p< b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™}½ö6 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼p<  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™}½ö6  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ééo�. bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1«ë€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¬,—‚míK‹ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ù ÙNNÃÃ)8*84­4­>$?$I™I™ST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÙÙNNÃà 8 8 ­ ­""——  ��öökkààU U"Ê"Ê$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1¿}!1п}!1]¾!1L]¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1‘ç!1(‘ç!1/("1¤/("1y/"1¬y/"19"1(9"19<"1ô9<"1cQ"1€cQ"1’"1ü’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1µÚ"1´ µÚ"1ÿ"10 ÿ"1W,#1PW,#1o2#1,o2#1k�#1˜k�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Î b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Zšó3 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Î b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Zšó3 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Î  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Zšó3  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 y™"w— bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€âbÄD¦&0p €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>-?-I¢I¢ST^Œ^Œhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbçbçd\e\gÑgÑiFjFl»l»n0o0q¥q¥stv�v�xy{y{y}î}îc€c‚؂؄M…M‡‡‰7Š7Œ¬Œ¬Ž!�!‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1~!1Ð~!1½¾!1L½¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1ñç!1(ñç!1�("1¤�("1y/"1¬y/"19"1(9"19<"1ô9<"1ÃQ"1€ÃQ"1a’"1üa’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1ûÙ"1´ ûÙ"1[þ"10 [þ"1�+#1P�+#1o2#1,o2#1‹Ž#1˜‹Ž#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1�&1Ì�&1áR'1áR'1‰e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯a- b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™™Ùó3 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯a- b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™™Ùó3 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1{¯a-  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™™Ùó3  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÉéOoÅå bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€övìlâbP� €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ä ä^^ÔÔ)J*J4À4À>6?6IªIªST^’^’hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjää^^ØØ R R Ì ÌFFÀÀ::´´..¨¨"" œ œ"#%Œ%Œ'(*x*x,î,î.d/d1Ú1Ú3P4P6Æ6Æ8<9<;²;²=(>(@ž@žBCEŠEŠGHJvJvLêLêN^O^QÒQÒSFTFVºVºX.Y.[¢[¢]^`Š`ŠbþbþdrergægæiZjZlÎlÎnBoBq¶q¶s*t*vžvžxy{†{†}ú}ún€n‚â‚â„V…V‡ʇʉ>Š>Œ²Œ²Ž&�&‘š‘š“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1¿}!1п}!1]¾!1L]¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1‘ç!1(‘ç!1/("1¤/("1y/"1¬y/"19"1(9"19<"1ô9<"1cQ"1€cQ"1’"1ü’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1©Ú"1´ ©Ú"1 ÿ"10  ÿ"1K,#1PK,#15#1,5#1 Ž#1˜ Ž#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1¥€&1Ì¥€&1uR'1uR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ì b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Zšó3 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ì b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Zšó3 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ì  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Zšó3  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 YyÛûlŒ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¢â€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ˆjêLÌ.®Ò €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÅÅ):*:4¯4¯>$?$I—I—S T ^}^}hðhðrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=ß=ß?T@TBÉBÉD>E>G³G³I(J(L›L›NOQ�Q�SôSôUgVgXÚXÚZM[M]À]À_3`3b¦b¦degŒgŒiÿiÿkrlrnånåpXqXsËsËu>v>x±x±z${$}—}— € ‚}‚}„ð„ð†c‡c‰Ö‰Ö‹IŒI޼޼�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1%~!1Ð%~!1þ!1Lþ!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1ñç!1(ñç!1�("1¤�("1y/"1¬y/"19"1(9"19<"1ô9<"1ÃQ"1€ÃQ"1a’"1üa’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"11Ú"1´ 1Ú"1‘þ"10 ‘þ"1Ó+#1PÓ+#1'5#1,'5#1?�#1˜?�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±e1 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™™Ùó3 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±e1 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™™Ùó3 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±e1  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™™Ùó3  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ééf†ü bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¶ö€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ØXÎNÄDº:F €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àXXÌÌ)@*@4´4´>(?(IšIšS T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  �� € €"ø"ø$l%l'à'à)T*T,È,È.Œ>ްް�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1Å}!1ÐÅ}!1c¾!1Lc¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1—ç!1(—ç!15("1¤5("1y/"1¬y/"19"1(9"19<"1ô9<"1iQ"1€iQ"1’"1ü’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1%Ú"1´ %Ú"1…þ"10 …þ"1Ç+#1PÇ+#1u2#1,u2#1Ž#1˜Ž#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1í€&1Ìí€&1½R'1½R'1ee'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ï b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Zšó3 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ï b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Zšó3 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ï  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Zšó3  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÉénŽì  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ímÚZÇGà  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á áWWÍÍ)C*C4¹4¹>-?-I¡I¡ST^‰^‰hýhýrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±''�� ‰ ‰"ÿ"ÿ$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@…@…BùBùDmEmGáGáIUJULÉLÉN=O=Q±Q±S%T%V™V™X Y [�[�]õ]õ_i`ibÝbÝdQeQgÅgÅi9j9l­l­n!o!q•q•s t v}v}xñxñze{e}Ù}ÙM€M‚Á‚Á„5…5‡©‡©‰ŠŒ�Œ�Ž�‘u‘u“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1~!1Ð~!1·¾!1L·¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1ëç!1(ëç!1‰("1¤‰("1y/"1¬y/"19"1(9"19<"1ô9<"1½Q"1€½Q"1[’"1ü[’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1ÁÚ"1´ ÁÚ"1!ÿ"10 !ÿ"1c,#1Pc,#1u2#1,u2#1?�#1˜?�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤V" b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™¿ò2 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤V" b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™¿ò2 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤V"  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™¿ò2  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Bb¯Ï�¡ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€�‚ƒ@ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝRRÇÇ)<*<4±4±>$?$I—I—S T ^}^}hðhðrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��wwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?P@PBÃBÃD6E6G©G©IJL�L�NOQuQuSèSèU[V[XÎXÎZA[A]´]´_'`'bšbšd e g€g€ióiókflfnÙnÙpLqLs¿s¿u2v2x¥x¥z{}‹}‹þþ�q‚q„ä„ä†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1Å}!1ÐÅ}!1c¾!1Lc¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1‘ç!1(‘ç!1/("1¤/("1y/"1¬y/"19"1(9"19<"1ô9<"1cQ"1€cQ"1’"1ü’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1Ú"1´ Ú"1aþ"10 aþ"1£+#1P£+#1!5#1,!5#1ÍŽ#1˜ÍŽ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1u€&1Ìu€&1ER'1ER'1íd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Í b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Zšó3 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Í b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Zšó3 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Í  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Zšó3  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÉéKkÆæ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1­í€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1´4¡!Ž{û�Í €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßSSÇÇ);*;4¯4¯>!?!I“I“ST^w^whéhér[s[} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ‡‡ûûo o"ã"ã$W%W'Ë'Ë)?*?,³,³.'/'1›1›346ƒ6ƒ8÷8÷:k;k=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b�b�dÿdÿfqgqiãiãkUlUnÇnÇp9q9s«s«uvx�x�z{}s}såå�W‚W„ɄɆ;‡;‰¬‰¬‹ŒŽŽŽŽ�ÿ�ÿ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1~!1Ð~!1½¾!1L½¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1ëç!1(ëç!1‰("1¤‰("1y/"1¬y/"19"1(9"19<"1ô9<"1½Q"1€½Q"1[’"1ü[’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1µÚ"1´ µÚ"1ÿ"10 ÿ"1W,#1PW,#135#1,35#1ÿ�#1˜ÿ�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%13€&1Ì3€&1R'1R'1«d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦Z& b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™~¾ò2 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦Z& b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™~¾ò2 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦Z&  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™~¾ò2  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ]}ê t” bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Á€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1„…†‡Á €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝPPÃÃ)6*64©4©>?I‹I‹SüSü]m^mhÞhÞrOsO} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��ssææY Y"Ì"Ì$?%?'²'²)%*%,˜,˜. / 1~1~3ñ3ñ5d6d8×8×:J;J=½=½?.@.BŸBŸDEG�G�IòIòKcLcNÔNÔPEQES¶S¶U'V'X˜X˜Z [ ]z]z_ë_ëa\b\dÍdÍf>g>i¯i¯k l n‘n‘pqssssuäuäwUxUzÆzÆ|7}7¨¨�‚„ЄІû†ûˆk‰k‹Û‹Û�KŽK�»�»’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1Ë}!1ÐË}!1i¾!1Li¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1�ç!1(�ç!1;("1¤;("1/"1¬/"1 9"1( 9"1?<"1ô?<"1oQ"1€oQ"1 ’"1ü ’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1Ú"1´ Ú"1sþ"10 sþ"1µ+#1Pµ+#1u2#1,u2#1ÓŽ#1˜ÓŽ#1ç$1$ç$1¿ò$1è¿ò$1ùÞ%1ìùÞ%1A�&1ÌA�&1S'1S'1¹e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ð b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @Zšó3 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ð b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @Zšó3 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ð  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @Zšó3  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÉécƒÖö bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€øxðpèh�Ð €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg à àYYÒÒ)K*K4Ä4Ä>=?=I´I´S)T)^œ^œhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggààYYÒÒ K K Ä Ä==¶¶//¨¨!!šš Œ Œ"#%~%~'÷'÷)p*p,é,é.b/b1Û1Û3T4T6Í6Í8F9F;¿;¿=8>8@±@±B*C*E£E£GHJ•J•L M OƒOƒQúQúSqTqVèVèX_Y_[Ö[Ö]M^M`Ä`Äb9c9e®e®g#h#j˜j˜l m o€o€qóqósftfvÙvÙxLyL{¿{¿}2~2€¥€¥‚ƒ…‹…‹‡þ‡þ‰pŠpŒâŒâŽT�T‘ƑƓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1~!1Ð~!1±¾!1L±¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1ßç!1(ßç!1}("1¤}("1/"1¬/"1 9"1( 9"1?<"1ô?<"1±Q"1€±Q"1O’"1üO’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1u2#1,u2#1Õ�#1˜Õ�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1á€&1Ìá€&1±R'1±R'1Ye'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™K b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™™Ùó3 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™K b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™™Ùó3 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™K  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™™Ùó3  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 5U™¹†¦ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ Š”ž�Ð €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âWWÌÌ)A*A4¶4¶>,?,I I ST^ˆ^ˆhühürrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢ŒŒ v v"ë"ë$`%`'Õ'Õ)J*J,¿,¿.4/41©1©346“6“89;};}=ò=ò?g@gBÜBÜDQEQGÆGÆI1?1I¦I¦ST^Š^Šhûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIYJYLÎLÎNCOCQ¸Q¸S-T-V¢V¢XY[Œ[Œ]^`v`vbébéd\e\gÏgÏiBjBlµlµn&o&q—q—stvyvyxêxêz[{[}Ì}Ì=€=‚®‚®„…‡‘‡‘‰ŠŒsŒsŽäŽä�U‘U“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1~!1Ð~!1±¾!1L±¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1åç!1(åç!1ƒ("1¤ƒ("1/"1¬/"1 9"1( 9"1?<"1ô?<"1·Q"1€·Q"1U’"1üU’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1?5#1,?5#1›�#1˜›�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1Ÿ€&1ÌŸ€&1oR'1oR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›O b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Y™™Ùó3 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›O b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Y™™Ùó3 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›O  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Y™™Ùó3  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 TtØø‡§ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ê €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1(¨2²<¼FÆZš €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞSSÈÈ)=*=4²4²>'?'IšIšS T ^~^~hðhðrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽxxííb b"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¤b¤degˆgˆiúiúklllnÞnÞpPqPsÂsÂu4v4x¦x¦z{}Š}Šüü�n‚n„à„à†T‡T‰ljNj:Œ:Ž­Ž­� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1ã}!1�ã}!1�¾!1�¾!1¿Â!1$¿Â!1µç!1�µç!1S("1S("1‘,"1$‘,"1‡Q"1�‡Q"1%’"1%’"1c–"1$c–"1Ù"1�Ù"1wý"1üwý"1;0#1 ;0#1#1| #1ÿ$1Ü ÿ$1×ò$1  ×ò$1ß%1¤ ß%1k�&1„ k�&1;S'1È;S'1ãe'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14q¥âÍ™ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14q¥âÍ™ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14q¥âÍ™ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Éé  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx ð ðhhàà)X*X4Ð4Ð>F?FI¼I¼S2T2^¨^¨his”s”} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðxxððhhàà X X Ð ÐHHÀÀ88°°((   � �"#%€%€'ø'ø)p*p,è,è.`/`1Ø1Ø3P4P6È6È8@9@;¸;¸=0>0@¨@¨B C E˜E˜GHJ†J†LüLüNrOrQèQèS^T^VÔVÔXJYJ[À[À]6^6`¬`¬b"c"e˜e˜ghj„j„lúlúnpopqæqæs\t\vÒvÒxHyH{¾{¾}4~4€ª€ª‚ ƒ …–…–‡ ˆ ЂЂŒøŒøŽn�n‘ä‘ä“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1 ~!1� ~!1«¾!1 «¾!1SÅ!1SÅ!1ßç!1€ßç!1}("1ü}("1%/"1%/"1±Q"1p±Q"1O’"1ìO’"1÷˜"1ô÷˜"1ÁÚ"1`ÁÚ"1!ÿ"1Ì!ÿ"1;0#1ì;0#1#1L #1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1�&1T �&1çR'1˜çR'1�e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�?  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1•Õ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�?  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1•Õ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�?  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1•Õ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1t”  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss ê êaaØØ)O*O4Æ4Æ>=?=I´I´S+T+^�^�his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssêêaaØØ O O Æ Æ==´´++¢¢�� ~ ~"õ"õ$l%l'ã'ã)Z*Z,Ñ,Ñ.H/H1¿1¿36466­6­8$9$;›;›=>@‰@‰BCEwEwGîGîIeJeLÜLÜNSOSQÊQÊSATAV¸V¸X/Y/[¦[¦]^`”`”b c e‚e‚gùgùipjplçlçnYoYqËqËs=t=v¯v¯x!y!{“{“}~€w€w‚é‚é„[…[‡͇͉?Š?Œ±Œ±Ž#�#‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1ã}!1�ã}!1�¾!1�¾!1¿Â!1$¿Â!1¯ç!1�¯ç!1M("1M("1‘,"1$‘,"1�Q"1��Q"1’"1’"1c–"1$c–"1Ù"1�Ù"1wý"1üwý"1;0#1 ;0#1#1| #1ç$1Ü ç$1¿ò$1  ¿ò$1ùÞ%1¤ ùÞ%1ó€&1„ ó€&1ÃR'1ÈÃR'1ke'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14o£ÞΚ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14o£ÞΚ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14o£ÞΚ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Éé  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ì ìbbØØ)N*N4Ä4Ä>8?8I¬I¬S T ^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvììbbØØ N N Ä Ä::°°&&œœˆˆþþt t"ê"ê$`%`'Ö'Ö)L*L,Â,Â.8/81®1®3$4$6š6š89;†;†=ü=ü?r@rBèBèD^E^GÔGÔIHJHL¼L¼N0O0Q¤Q¤STVŒVŒXY[t[t]è]è_\`\bÐbÐdDeDg¸g¸i,j,l l noqˆqˆsüsüupvpxäxäzX{X}Ì}Ì@€@‚´‚´„(…(‡œ‡œ‰ŠŒ„Œ„ŽøŽø�l‘l“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1 ~!1� ~!1«¾!1 «¾!1SÅ!1SÅ!1ßç!1€ßç!1}("1ü}("1%/"1%/"1±Q"1p±Q"1O’"1ìO’"1÷˜"1ô÷˜"1ÍÚ"1`ÍÚ"1-ÿ"1Ì-ÿ"1;0#1ì;0#1#1L #1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1À&1T À&1“R'1˜“R'1;e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�A  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1”Ô€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�A  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1”Ô€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1Y�A  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1”Ô €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ÿ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss è è]]ÒÒ)G*G4¼4¼>1?1I¦I¦ST^�^�hÿhÿrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssèè]]ÒÒ G G ¼ ¼11¦¦��zzïïd d"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I)J)LžLžNOQˆQˆSýSýUrVrXçXçZ\[\]Ñ]Ñ_F`Fb»b»d0e0g¥g¥ijl�l�noqsqssåsåuWvWxÉxÉz;{;}­}­€‚‘‚‘„…‡u‡u‰ç‰ç‹YŒYŽËŽË�=‘=“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Ý}!1�Ý}!1{¾!1 {¾!1¹Â!1¹Â!1¯ç!1€¯ç!1M("1üM("1‹,"1‹,"1�Q"1p�Q"1’"1ì’"1]–"1ô]–"1Ù"1`Ù"1qý"1Ìqý"150#1ì50#1#1L #1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1#�&1T #�&1óR'1˜óR'1›e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢ß b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢ß b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢ß b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Éé  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ë ëbbÙÙ)P*P4Ç4Ç>>µµ,,££‘‘  "ö"ö$m%m'ä'ä)[*[,Ò,Ò.I/I1À1À37476®6®8%9%;œ;œ=>@Š@ŠBCExExGïGïIdJdLÙLÙNNONQÃQÃS8T8V­V­X"Y"[—[—] ^ `�`�böbödkekgàgàiUjUlÊlÊn?o?q´q´s)t)vžvžxy{ˆ{ˆ}ý}ýr€r‚ç‚ç„\…\‡чщFŠFŒ»Œ»Ž0�0‘¥‘¥“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1±¾!1 ±¾!1SÅ!1SÅ!1åç!1€åç!1ƒ("1üƒ("1%/"1%/"1·Q"1p·Q"1U’"1ìU’"1÷˜"1ô÷˜"1µÚ"1`µÚ"1ÿ"1Ìÿ"150#1ì50#1#1L #1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1#�&1T #�&1óR'1˜óR'1›e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–H b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1™Ù€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–H b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1™Ù€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–H b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1™Ù €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1o�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss í íggÝÝ)S*S4É4É>=?=I±I±S%T%^™^™h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssííggáá [ [ Õ ÕOOÉÉCC½½77±±++ ¥ ¥"#%•%•' ( *�*�,÷,÷.m/m1ã1ã3Y4Y6Ï6Ï8E9E;»;»=1>1@§@§BCE“E“G H J}J}LñLñNeOeQÙQÙSMTMVÁVÁX5Y5[©[©]^`‘`‘bceyeygígíiajalÕlÕnIoIq½q½s1t1v¥v¥xy{�{�}~€u€u‚é‚é„]…]‡чщEŠEŒ¹Œ¹Ž-�-‘¡‘¡“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1×}!1�×}!1u¾!1 u¾!1¹Â!1¹Â!1©ç!1€©ç!1G("1üG("1‹,"1‹,"1{Q"1p{Q"1’"1ì’"1]–"1ô]–"1Ù"1`Ù"1qý"1Ìqý"150#1ì50#1#1L #1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1½€&1T ½€&1�R'1˜�R'15e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l Û b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l Û b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l Û b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Éé  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss é é__ÔÔ)I*I4¾4¾>1?1I¤I¤ST^Š^Šhýhýrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðsséé__ÕÕ K K Á Á77­­##™™……ûûq q"ç"ç$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI5J5L¨L¨NOQŽQŽSTVtVtXçXçZZ[Z]Í]Í_@`@b³b³d&e&g™g™i j llnònòpeqesØsØuKvKx¾x¾z1{1}¤}¤€‚ЂЄý„ý†p‡p‰ã‰ã‹VŒVŽÉŽÉ�<‘<“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1±¾!1 ±¾!1SÅ!1SÅ!1åç!1€åç!1ƒ("1üƒ("1%/"1%/"1·Q"1p·Q"1U’"1ìU’"1÷˜"1ô÷˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"150#1ì50#1#1L #1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–J b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1™Ù€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–J b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1™Ù€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–J b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1™Ù €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1~ž  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss ë ëcc××)K*K4¿4¿>1?1I£I£ST^‡^‡hùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssëëccÛÛ S S Ë ËCC»»33««##›› ‹ ‹"#%w%w'ë'ë)_*_,Ó,Ó.G/G1»1»3/4/6£6£89;‹;‹=ÿ=ÿ?s@sBçBçD[E[GÏGÏIAJAL³L³N%O%Q—Q—S T V{V{XíXíZ_[_]Ñ]Ñ_C`Cbµbµd'e'g™g™i j l}l}nïnïpaqasÓsÓuEvEx·x·z){)}›}› € ‚‚„ñ„ñ†c‡c‰Õ‰Õ‹GŒG޹޹�+‘+“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1×}!1�×}!1u¾!1 u¾!1¿Â!1¿Â!1©ç!1€©ç!1G("1üG("1‘,"1‘,"1{Q"1p{Q"1’"1ì’"1c–"1ôc–"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1ù€&1T ù€&1ÉR'1˜ÉR'1qe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸÜ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸÜ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸÜ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1k‹  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt î îddÚÚ)P*P4Ä4Ä>8?8I¬I¬S T ^”^”hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttîîhhââ \ \ Ö ÖPPÊÊDD¾¾44ªª  – –" # %‚%‚'ø'ø)n*n,ä,ä.Z/Z1Ð1Ð3F4F6¼6¼8292;¨;¨=>@�@�BCExExGìGìI`J`LÔLÔNHOHQ¼Q¼S0T0V¤V¤XY[Œ[Œ]^`t`tbèbèd\e\gÐgÐiDjDl¸l¸n,o,q q stvˆvˆxüxüzp{p}ä}äX€X‚̂̄@…@‡²‡²‰%Š%Œ˜Œ˜Ž � ‘~‘~“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1·¾!1 ·¾!15Å!15Å!1åç!1€åç!1ƒ("1üƒ("1/"1/"1·Q"1p·Q"1U’"1ìU’"1Ù˜"1ôÙ˜"1 Ù"1` Ù"1ký"1Ìký"1/0#1ì/0#1û~#1L û~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸQ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1}½€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸQ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1}½€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸQ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1}½ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Éé  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll â âWWÌÌ)A*A4´4´>'?'IšIšS T ^€^€hóhórdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllââXXÎÎ D D º º00¦¦’’||ññf f"Û"Û$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=à=à?S@SBÆBÆD9E9G¬G¬IJL’L’NOQxQxSëSëU^V^XÑXÑZD[D]·]·_*`*b�b�degƒgƒiöiökilinÜnÜpOqOsÂsÂu5v5x¨x¨z{}Ž}Ž€‚t‚t„ç„ç†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ñ}!1�Ñ}!1o¾!1 o¾!1¿Â!1¿Â!1£ç!1€£ç!1A("1üA("1‘,"1‘,"1uQ"1puQ"1’"1ì’"1c–"1ôc–"1µÚ"1`µÚ"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i� Ø b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i� Ø b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i� Ø b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1p�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ê ê^^ÒÒ)F*F4¸4¸>*?*IœIœST^€^€hòhòrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrêêbbÚÚ R R Ê ÊBBºº22ªª’’ z z"î"î$b%b'Ö'Ö)J*J,¾,¾.2/21¦1¦346Ž6Ž89;v;v=è=è?Z@ZBÌBÌD>E>G°G°I"J"L”L”NOQxQxSêSêU\V\XÎXÎZ@[@]²]²_$`$b–b–degzgziìiìk^l^nÐnÐpBqBs´s´u&v&x˜x˜z { }|}|îî�`‚`„Ò„Ò†B‡B‰³‰³‹$Œ$ޕޕ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1·¾!1 ·¾!15Å!15Å!1ëç!1€ëç!1‰("1ü‰("1/"1/"1½Q"1p½Q"1[’"1ì[’"1Ù˜"1ôÙ˜"1 Ù"1` Ù"1ký"1Ìký"1/0#1ì/0#1û~#1L û~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%19€&1T 9€&1 R'1˜ R'1±d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸS b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1}½€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸS b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1}½€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸS b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1}½ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Éé  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn ä äWWÊÊ)=*=4®4®>?I�I�ST^r^rhãhãrRsR} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnääZZÐÐ F F ¼ ¼22¨¨””zzíí` `"Ó"Ó$F%F'¹'¹),*,,Ÿ,Ÿ./1…1…3ø3ø5k6k8Þ8Þ:Q;Q=Â=Â?3@3B¤B¤DEG†G†I÷I÷KhLhNÙNÙPJQJS»S»U,V,X�X�Z[]]_ð_ðaabadÒdÒfCgCi´i´k%l%n–n–pqsxsxuéuéwZxZzËzË|<}<­­�‚„�„�†þ†þˆn‰n‹Þ‹Þ�NŽN�¾�¾’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ñ}!1�Ñ}!1o¾!1 o¾!1³Â!1³Â!1�ç!1€�ç!1;("1ü;("1…,"1…,"1oQ"1poQ"1 ’"1ì ’"1W–"1ôW–"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1/0#1ì/0#1û~#1L û~#1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1A�&1T A�&1S'1˜S'1¹e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœ Ù b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœ Ù b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœ Ù b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1f†  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp é ébbÛÛ)T*T4Í4Í>D?DI»I»S0T0^£^£his‡s‡} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppéébbÛÛ T T Í ÍFF¿¿88±±**££ • •"#%‡%‡'(*y*y,ò,ò.k/k1ä1ä3]4]6Ö6Ö8O9O;È;È=A>A@º@ºB3C3E¬E¬G%H%JœJœLMOŠOŠQRTxTxVïVïXfYf[Ý[Ý]T^T`Ë`Ëb@c@eµeµg*h*jŸjŸlmo‡o‡qúqúsmtmvàvàxSyS{Æ{Æ}9~9€¬€¬‚ƒ…’…’‡ˆŠuŠuŒçŒçŽY�Y‘ˑ˓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1½¾!1 ½¾!1SÅ!1SÅ!1ëç!1€ëç!1‰("1ü‰("1%/"1%/"1½Q"1p½Q"1[’"1ì[’"1÷˜"1ô÷˜"1 Ù"1` Ù"1ký"1Ìký"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨Z& b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¢â€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨Z& b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¢â€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨Z& b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¢â €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Éé  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk ã ãXXÍÍ)B*B4·4·>+?+IŸIŸST^‡^‡hûhûroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkãã[[ÓÓ K K à Ã;;³³++££�� w w"ì"ì$a%a'Ö'Ö)K*K,À,À.5/51ª1ª346”6”8 9 ;~;~=ó=ó?h@hBÝBÝDRERGÇGÇI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÃbÃd7e7g«g«ijl“l“noq{q{sïsïucvcx×x×zK{K}¿}¿3€3‚§‚§„…‡�‡�‰ŠŒyŒyŽîŽî�c‘c“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ë}!1�Ë}!1i¾!1 i¾!1³Â!1³Â!1�ç!1€�ç!1;("1ü;("1…,"1…,"1oQ"1poQ"1 ’"1ì ’"1W–"1ôW–"1»Ú"1`»Ú"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Ï€&1T Ï€&1ŸR'1˜ŸR'1Ge'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš Õ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš Õ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fš Õ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1}�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn å å\\ÓÓ)J*J4Á4Á>6?6I«I«ST^�^�hispsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnåå\\ÓÓ J J Á Á88¯¯&&��‹‹ y y"ð"ð$g%g'Þ'Þ)U*U,Ì,Ì.C/C1º1º31416¨6¨89;–;–= > @„@„BûBûDrErGéGéI^J^LÓLÓNHOHQ½Q½S2T2V§V§XY[‘[‘]^`{`{bîbîdaeagÔgÔiGjGlºlºn+o+qœqœs t v~v~xïxïz`{`}Ñ}ÑB€B‚³‚³„$…$‡”‡”‰ŠŒvŒvŽçŽç�X‘X“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1½¾!1 ½¾!1SÅ!1SÅ!1ñç!1€ñç!1�("1ü�("1%/"1%/"1ÃQ"1pÃQ"1a’"1ìa’"1÷˜"1ô÷˜"1 Ù"1` Ù"1ký"1Ìký"1/0#1ì/0#1û~#1L û~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨\( b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1£ã€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨\( b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1£ã€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨\( b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1£ã €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Éé  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk á áVVËË)@*@4µ4µ>(?(I›I›S T ^^hñhñrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkááWWÍÍ C C ¹ ¹//¥¥‘‘{{ððe e"Ú"Ú$O%O'Ä'Ä)9*9,®,®.#/#1˜1˜3 4 6‚6‚8÷8÷:l;l=á=á?V@VBËBËD@E@GµGµI(J(L›L›NOQ�Q�SôSôUgVgXÚXÚZM[M]À]À_3`3b¥b¥deg‰g‰iûiûkmlmnßnßpQqQsÃsÃu5v5x§x§z{}‹}‹ýý�o‚o„á„á†S‡S‰ƉƋ9Œ9ެެ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ë}!1�Ë}!1i¾!1 i¾!1¿Â!1¿Â!1—ç!1€—ç!15("1ü5("1‘,"1‘,"1iQ"1piQ"1’"1ì’"1c–"1ôc–"1�Ú"1`�Ú"1ýþ"1Ìýþ"1)0#1ì)0#1õ~#1L õ~#1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1M�&1T M�&1S'1˜S'1Åe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™ Ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™ Ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™ Ö b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1a�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ä ä\\ÔÔ)L*L4Ä4Ä>:?:I°I°S&T&^œ^œhisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllää\\ÔÔ L L Ä Ä<<´´,,¤¤””  „ „"ü"ü$t%t'ì'ì)d*d,Ü,Ü.T/T1Ì1Ì3D4D6¼6¼8494;¬;¬=$>$@œ@œBCEŒEŒGHJzJzLðLðNfOfQÜQÜSRTRVÈVÈX>Y>[´[´]*^*` ` bceŒeŒghjxjxlîlîndodqÚqÚsPtPvÆvÆx1?1I¨I¨ST^‘^‘hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIYJYLÐLÐNGOGQ¾Q¾S5T5V¬V¬X#Y#[š[š]^`ˆ`ˆbÿbÿdvevgígíidjdlÛlÛnMoMq¿q¿s1t1v£v£xy{‡{‡}ù}ùk€k‚݂݄O…O‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Å}!1�Å}!1c¾!1 c¾!1¿Â!1¿Â!1—ç!1€—ç!15("1ü5("1‘,"1‘,"1iQ"1piQ"1’"1ì’"1c–"1ôc–"1»Ú"1`»Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Õ€&1T Õ€&1¥R'1˜¥R'1Me'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ò b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ò b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ò b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1€   bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj à àVVÌÌ)B*B4¸4¸>,?,I I ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjààVVÌÌ B B ¸ ¸..¤¤��||òòh h"Þ"Þ$T%T'Ê'Ê)@*@,¶,¶.,/,1¢1¢346Ž6Ž89;z;z=ð=ð?f@fBÜBÜDRERGÈGÈI%?%IšIšST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÜÜQQÆÆ ; ; ° °%%šš„„ùùnnããX X"Í"Í$B%B'·'·),*,,¡,¡./1‹1‹346u6u8ê8ê:_;_=Ô=Ô?I@IB¾B¾D3E3G¨G¨IJL’L’NOQ|Q|SñSñUfVfXÛXÛZP[P]Å]Å_:`:b¯b¯d$e$g™g™ijlƒlƒnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¿}!1�¿}!1]¾!1 ]¾!1§Â!1§Â!1‘ç!1€‘ç!1/("1ü/("1y,"1y,"1cQ"1pcQ"1’"1ì’"1K–"1ôK–"1‘Ú"1`‘Ú"1ñþ"1Ìñþ"1#0#1ì#0#1ï~#1L ï~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1 �&1T  �&1ÛR'1˜ÛR'1ƒe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–Ó b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–Ó b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–Ó b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1[{  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>0?0I¥I¥ST^�^�hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubêbêd_e_gÔgÔiIjIl¾l¾n3o3q¨q¨stv’v’xy{|{|}ñ}ñf€f‚Û‚Û„P…P‡Ňʼn:Š:Œ¯Œ¯Ž$�$‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1%~!1�%~!1þ!1 þ!1GÅ!1GÅ!1÷ç!1€÷ç!1•("1ü•("1/"1/"1ÉQ"1pÉQ"1g’"1ìg’"1ë˜"1ôë˜"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1#0#1ì#0#1ï~#1L ï~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºl8 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1žÞ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºl8 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1žÞ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºl8 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1žÞ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Éé  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á á[[ÑÑ)G*G4½4½>1?1I¥I¥ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±++¥¥ ™ ™"#%‰%‰'ÿ'ÿ)u*u,ë,ë.a/a1×1×3M4M6Ã6Ã8999;¯;¯=%>%@›@›BCE‡E‡GýGýIqJqLåLåNYOYQÍQÍSATAVµVµX)Y)[�[�]^`…`…bùbùdmemgágáiUjUlÉlÉn=o=q±q±s%t%v™v™x y {�{�}õ}õi€i‚݂݄Q…Q‡Ňʼn9Š9Œ­Œ­Ž!�!‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¿}!1�¿}!1]¾!1 ]¾!1§Â!1§Â!1‹ç!1€‹ç!1)("1ü)("1y,"1y,"1]Q"1p]Q"1û‘"1ìû‘"1K–"1ôK–"1»Ú"1`»Ú"1ÿ"1Ìÿ"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ï b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ï b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ï b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1�¡  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝSSÈÈ)=*=4²4²>%?%I˜I˜S T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��yyïïe e"Û"Û$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnænæpYqYsÌsÌu?v?x²x²z%{%}˜}˜ € ‚~‚~„ñ„ñ†d‡d‰׉׋JŒJ޽޽�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1+~!1�+~!1ɾ!1 ɾ!1GÅ!1GÅ!1÷ç!1€÷ç!1•("1ü•("1/"1/"1ÉQ"1pÉQ"1g’"1ìg’"1ë˜"1ôë˜"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºn: b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ÿß€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºn: b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ÿß€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1†ºn: b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ÿß €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Éé  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßWWËË)?*?4³4³>%?%I—I—S T ^{^{híhír_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ��  "÷"÷$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnãnãpUqUsÇsÇu9v9x«x«z{}�}�€‚s‚s„å„å†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¹}!1�¹}!1W¾!1 W¾!1³Â!1³Â!1‹ç!1€‹ç!1)("1ü)("1…,"1…,"1]Q"1p]Q"1û‘"1ìû‘"1W–"1ôW–"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1á€&1T á€&1±R'1˜±R'1Ye'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ð b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ð b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ð b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Éé  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh â âXXÎÎ)D*D4¸4¸>,?,I I ST^Š^Šhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²((žž Š Š"#%v%v'ì'ì)b*b,Ø,Ø.N/N1Ä1Ä3:4:6°6°8&9&;œ;œ=>@„@„BøBøDlElGàGàITJTLÈLÈN?IŽIŽS T ^~^~hñhñrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††ûûppååZ Z"Ï"Ï$D%D'¹'¹).*.,£,£./1�1�346w6w8ì8ì:a;a=Ô=Ô?G@GBºBºD-E-G G IJL†L†NùNùPlQlSßSßURVRXÅXÅZ8[8]«]«_`b›b›deg�g�iôiôkglgnÚnÚpMqMsÀsÀu3v3x¦x¦z{}Œ}Œÿÿ�r‚r„å„å†V‡V‰ȉȋ:Œ:ެެ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¹}!1�¹}!1W¾!1 W¾!1³Â!1³Â!1…ç!1€…ç!1#("1ü#("1…,"1…,"1WQ"1pWQ"1õ‘"1ìõ‘"1W–"1ôW–"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1#0#1ì#0#1ï~#1L ï~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1i€&1T i€&19R'1˜9R'1ád'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ì b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ì b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ì b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Éé  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞRRÆÆ):*:4¬4¬>?I�I�ST^x^xhêhêrZsZ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žž††úún n"â"â$V%V'Ê'Ê)>*>,²,².&/&1š1š346‚6‚8ö8ö:j;j=Ü=Ü?N@NBÀBÀD2E2G¤G¤IJLˆLˆNúNúPlQlSÞSÞUPVPXÂXÂZ4[4]¦]¦_`bŽbŽdegrgriäiäkVlVnÈnÈp:q:s¬s¬uvx�x�z{}t}tææ�X‚X„ʄʆ:‡:‰«‰«‹ŒŽ�Ž��þ�þ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1+~!1�+~!1ɾ!1 ɾ!1#Å!1#Å!1ýç!1€ýç!1›("1ü›("1õ."1õ."1ÏQ"1pÏQ"1m’"1ìm’"1ǘ"1ôǘ"1—Ú"1`—Ú"1÷þ"1Ì÷þ"1#0#1ì#0#1ï~#1L ï~#1¥$1¬ ¥$1}ò$1p }ò$1·Þ%1t ·Þ%1'€&1T '€&1÷Q'1˜÷Q'1Ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�ÃwC b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1}½€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�ÃwC b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1}½€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�ÃwC b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1}½ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1a�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØKK¾¾)1*14¢4¢>?I„I„SýSý]n^nhßhßrNsN} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆûûnnááT T"Ç"Ç$:%:'­'­) * ,“,“./1y1y3ì3ì5_6_8Ò8Ò:E;E=¶=¶?'@'B˜B˜D E GzGzIëIëK\L\NÍNÍP>Q>S¯S¯U V X‘X‘Z[]s]s_ä_äa]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1³}!1�³}!1Q¾!1 Q¾!1¡Â!1¡Â!1…ç!1€…ç!1#("1ü#("1s,"1s,"1WQ"1pWQ"1õ‘"1ìõ‘"1E–"1ôE–"11Ú"1`1Ú"1‘þ"1Ô‘þ"10#1ô0#1é~#1T é~#1Û$1´ Û$1ñ$1x ñ$1áÀ%1| áÀ%15�&1\ 5�&1+Q'1 +Q'1Óc'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�Í b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�Í b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�Í b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1ÃS³  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ý ÝVVÑÑ)J*J4Ã4Ã>:?:I±I±S&T&^¥^¥his‰s‰} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÝÝVVÏÏ H H Á Á::³³,,¥¥—— ‰ ‰"#%}%}'ö'ö)o*o,è,è.a/a1Ú1Ú3S4S6Ì6Ì8E9E;¾;¾=7>7@°@°B)C)E¢E¢GHJ’J’L M O€O€Q÷Q÷SnTnVåVåX\Y\[Ó[Ó]J^J`Á`Áb6c6e«e«g h j•j•l m o‰o‰qüqüsotovâvâxUyU{È{È};~;€®€®‚!ƒ!…”…”‡ˆŠwŠwŒéŒéŽ[�[‘͓͑ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!11~!1�1~!1Ͼ!1 Ͼ!1;Å!1;Å!1è!1€è!1¡("1ü¡("1 /"1 /"1ÕQ"1pÕQ"1s’"1ìs’"1ߘ"1ôߘ"1ÕØ"1`ÕØ"15ý"1Ì5ý"10#1ì0#1é~#1L é~#1É$1¬ É$1ñð$1p ñð$1ÏÀ%1t ÏÀ%1Õ€&1T Õ€&1ËP'1˜ËP'1sc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜Ì~J b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1œÜ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜Ì~J b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1œÜ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜Ì~J b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1œÜ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ÉÅ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ × ×LLËË)@*@4µ4µ>)?)I�I�ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__××OOÇÇ ? ? · ·//§§——  ��öök k"à"à$_%_'Ô'Ô)I*I,¾,¾.3/31¨1¨346’6’89;|;|=ñ=ñ?f@fBÛBÛDPEPGÅGÅI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÍbÍdAeAgµgµi)j)l�l�noq…q…sùsùumvmxáxázU{U}É}É=€=‚±‚±„%…%‡™‡™‰ŠŒƒŒƒŽøŽø�m‘m“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!1­}!1�­}!1K¾!1 K¾!1¡Â!1¡Â!1ç!1€ç!1("1ü("1s,"1s,"1QQ"1pQQ"1ï‘"1ìï‘"1E–"1ôE–"1IÚ"1`IÚ"1©þ"1Ô©þ"10#1ô0#1é~#1T é~#1Ã$1´ Ã$1ëð$1x ëð$1ÉÀ%1| ÉÀ%1À&1\ À&1¹P'1 ¹P'1ac'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽýÉ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽýÉ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ZŽýÉ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1áqÊ6  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPËË)B*B4¹4¹>.?.I£I£ST^“^“histst} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbæbædYeYgÌgÌi?j?l²l²n/o/q q stv‚v‚xóxózd{d}Õ}ÕF€F‚·‚·„(…(‡˜‡˜‰ Š ŒzŒzŽëŽë�\‘\“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!11~!1�1~!1Ͼ!1 Ͼ!1AÅ!1AÅ!1è!1€è!1¡("1ü¡("1/"1/"1ÕQ"1pÕQ"1s’"1ìs’"1å˜"1ôå˜"1ÕØ"1`ÕØ"15ý"1Ì5ý"10#1ì0#1é~#1L é~#1½$1¬ ½$1åð$1p åð$1ÃÀ%1t ÃÀ%1�€&1T �€&1ƒP'1˜ƒP'1+c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜Ì€L b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ý€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜Ì€L b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ý€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1˜Ì€L b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ý €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ÉÅ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Õ ÕJJÇÇ)<*<4±4±>$?$I—I—ST^‡^‡hùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……úúooääY Y"Î"Î$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇDA�Ü bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ Õ ÕÍÍ)D*D4»4»>4?4I«I«S"T"^ ^ his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÕÕLLÃà : : ± ±((ŸŸ��ÆÆ==´´+ +"¢"¢$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãI\J\LÓLÓNJOJQÁQÁS8T8V¯V¯X&Y&[�[�]^`‹`‹bceyeygðgðigjglÞlÞn\o\qÎqÎs@t@v²v²x$y${–{–}~€z€z‚ì‚ì„^…^‡ЇЉBŠBŒ´Œ´Ž&�&‘˜‘˜“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1I~!1èI~!1ç¾!1dç¾!1•Å!1l•Å!1Ï!1èÏ!1UÒ!1´UÒ!1è!1@è!1³("1¼³("1g/"1Äg/"1ñ8"1@ñ8"1'<"1 '<"1çQ"1˜çQ"1…’"1…’"19™"1 9™"1â"1˜ â"1ù¥"1d ù¥"1;Ó"1ð ;Ó"1¯Ú"1Ì ¯Ú"1ÿ"1H ÿ"1Q,#1hQ,#1‹:#1D‹:#1Go#1ÐGo#1 �#1ô �#1š#1 š#1Û$1¬Û$1³ò$1p³ò$1íÞ%1tíÞ%1gì%1Tgì%1A�&1àA�&1S'1$S'1¹e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1´èœh b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [›i©ö6 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1´èœh b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [›i©ö6 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1´èœh  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [›i©ö6  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 r’4ˆ¨ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1[›ö6N‘Ñ,l  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d[Û6¶‘ìlÇG¢"}ýXØ,¬3³ñq¶6Ž{ûéi Ä D Ÿ  z ú U Õ 0 ° ^ ž  K¦æA�Ü bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSS��)E*E4½4½>5?5I«I«S!T!^£^£his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$/%/'§'§)*,—,—./1Å1Å3=4=6µ6µ8-9-;¥;¥=>@•@•B C E…E…GýGýIuJuLëLëNaOaQ×Q×SMTMVÃVÃX9Y9[¯[¯]%^%`›`›bce‡e‡gýgýisjslélénkokqáqásWtWvÍvÍxCyC{¹{¹}/~/€¥€¥‚ƒ…‘…‘‡ˆŠ}Š}ŒóŒóŽi�i‘ߑߓ  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„1i!1e}!1e}!1¾!1|¾!1mÃ!1„mÃ!1£Æ!1ð£Æ!1-Ð!1|-Ð!1cÓ!1HcÓ!17ç!1Ô7ç!1Õ'"1PÕ'"1?-"1X?-"1u0"1Äu0"1ÿ9"1Pÿ9"15="15="1 Q"1¨ Q"1§‘"1$ §‘"1—"1, —"1Gš"1˜ Gš"1Ñ£"1$ Ñ£"1§"1ð §"1IÔ"1| IÔ"1CÚ"1XCÚ"1£þ"1Ô£þ"1å+#1ôå+#1�:#1Ð�:#1Yo#1\Yo#1W�#1€W�#1Uš#1¬Uš#1½$18½$1•ò$1ü•ò$1ÏÞ%1ÏÞ%1Iì%1àIì%1Ÿ€&1lŸ€&1oR'1°oR'1e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OÀŒ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ë  bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OÀŒ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ë   bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OÀŒ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ë  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ë뇧> bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1` @žÞé) <|  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d`à@À  €à`À@  €^Þ<¼ š ø x \ Ü Ö V 7 · ´ 4  ’ ’  í m p ð P�ð0ŽÎ™Ù,l bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ Ó Ó  ÅÅ):*:4¯4¯>&?&I›I›ST^Ž^Žhisrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÓÓHH½½ 2 2 § §‘‘{{±±&&›› "…"…$A%A'¶'¶)+*+, , ./1Š1Š3ÿ3ÿ5t6t8é8é:^;^=Ó=Ó?H@HB½B½D2E2G§G§IJL“L“NOQ}Q}SòSòUgVgXÜXÜZQ[Q]Æ]Æ_;`;b°b°d%e%gšgšijl„l„noqtqtsæsæuXvXxÊxÊz<{<}®}® € ‚’‚’„…‡v‡v‰è‰è‹ZŒZŽÌŽÌ�>‘>“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1I~!1èI~!1ç¾!1dç¾!1•Å!1l•Å!1Ï!1èÏ!1UÒ!1´UÒ!1è!1@è!1¹("1¼¹("1g/"1Äg/"1ñ8"1@ñ8"1'<"1 '<"1íQ"1˜íQ"1‹’"1‹’"19™"1 9™"1â"1˜ â"1ù¥"1d ù¥"1;Ó"1ð ;Ó"1Ú"1Ì Ú"1ßþ"1H ßþ"1!,#1h!,#1‹:#1D‹:#1Go#1ÐGo#1 �#1ô �#1š#1 š#1Ã$1¬Ã$1›ò$1p›ò$1ÕÞ%1tÕÞ%1Oì%1TOì%1É€&1àÉ€&1™R'1$™R'1Ae'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¶ê l b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [›jªö6 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¶ê l b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [›jªö6 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¶ê l  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [›jªö6  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ’²Ww<\ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1[›ö6‘Ñ“Ó ,l  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d[Û6¶‘ìlÇG¢"}ýXØ3³Žéi @ À Ä D  … Ÿ  Ê J z ú �  U Õ 0 °  Kã #¦æA�Ü bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×MMˆˆ)=*=4³4³>)?)I�I�ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$%'†'†)ü)ü+r,r.è.è0�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noq…q…sùsùumvmxáxázU{U}É}É=€=‚±‚±„%…%‡™‡™‰ Š Œ�Œ�ŽõŽõ�i‘i“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx7i!1³}!1è³}!1Q¾!1dQ¾!1›Å!1l›Å!1%Ï!1è%Ï!1[Ò!1´[Ò!1ç!1@ç!1("1¼("1m/"1Äm/"1÷8"1@÷8"1-<"1 -<"1QQ"1˜QQ"1ï‘"1ï‘"1?™"1 ?™"1É¢"1˜ É¢"1ÿ¥"1d ÿ¥"1AÓ"1ð AÓ"1CÚ"1Ì CÚ"1£þ"1H £þ"1å+#1hå+#1‘:#1D‘:#1Mo#1ÐMo#1�#1ô�#1 š#1  š#1É$1¬É$1¡ò$1p¡ò$1ÛÞ%1tÛÞ%1Uì%1TUì%1ÿ€&1àÿ€&1ÏR'1$ÏR'1we'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�Î b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\œ÷7 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�Î b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\œ÷7 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�Î  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\œ÷7  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ëëv–< bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1[›ö6I‰‘Ñ,l  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d[Û6¶‘ìlÇG¢"}ýXØß_3³¸8Ž‘ éi j ê Æ F ¡ ! | ü W × 2 ²  M¨èÙCƒÞ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ü ÜVVÌÌ)B*B4¸4¸>.?.I¢I¢SØSØ]”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÜÜVVÐÐ J J Ä Ä>>¸¸22¬¬&&   ” ”"#%„%„'ú'ú)p*p,æ,æ.\/\1Ò1Ò3H4H6¾6¾8494;ª;ª= > @–@–B C E‚E‚GøGøInJnLâLâNVOVQÊQÊS>T>V²V²X&Y&[š[š]^`‚`‚b¸b¸d,e,g g ijlˆlˆnDoDq¸q¸s,t,v v xy{ˆ{ˆ}ü}üp€p‚ä‚ä„X…X‡̇̉@Š@Œ´Œ´Ž(�(‘œ‘œ“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx7i!1C~!1èC~!1á¾!1dá¾!1›Å!1l›Å!1%Ï!1è%Ï!1[Ò!1´[Ò!1è!1@è!1­("1¼­("1m/"1Äm/"1÷8"1@÷8"1-<"1 -<"1áQ"1˜áQ"1’"1’"1?™"1 ?™"1É¢"1˜ É¢"1ÿ¥"1d ÿ¥"1AÓ"1ð AÓ"1ÇÚ"1Ì ÇÚ"1'ÿ"1H 'ÿ"1i,#1hi,#1‘:#1D‘:#1Mo#1ÐMo#1�#1ô�#1 š#1  š#1É$1¬É$1¡ò$1p¡ò$1ÛÞ%1tÛÞ%1Uì%1TUì%1ÿ€&1àÿ€&1ÏR'1$ÏR'1we'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1©Ý‘] b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [›­íö6 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1©Ý‘] b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [›­íö6 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1©Ý‘]  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [›­íö6  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u•:—· bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1[›ö6\‘Ñ,l  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d[Û6¶‘ìlÇG¢"}ýXØpð3³<¼Žˆéi ÔT Ä D Ÿ  z ú U Õ 0 °  Kì ,¦æA�Ü bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>-?-I¢I¢ST^P^Phisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbçbçd\e\gÑgÑiFjFl»l»nônôpiqisÞsÞuSvSxÈxÈz{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx7i!1­}!1è­}!1K¾!1dK¾!1›Å!1l›Å!1%Ï!1è%Ï!1[Ò!1´[Ò!1ç!1@ç!1("1¼("1m/"1Äm/"1÷8"1@÷8"1-<"1 -<"1QQ"1˜QQ"1ï‘"1ï‘"1?™"1 ?™"1É¢"1˜ É¢"1ÿ¥"1d ÿ¥"1AÓ"1ð AÓ"1IÚ"1Ì IÚ"1©þ"1H ©þ"1ë+#1hë+#1‘:#1D‘:#1Mo#1ÐMo#1�#1ô�#1 š#1  š#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1‡€&1à‡€&1WR'1$WR'1ÿd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�Ì b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\œ÷7 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�Ì b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\œ÷7 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�Ì  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\œ÷7  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ë눨 @ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1[›ö6‘Ñâ" ,l  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d[Û6¶‘ìlÇG¢"}ýXØ3³Žéi Ä D C à Ÿ   œ z ú õ u U Õ Î N 0 °  K¦æA�r²Ü bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚRRÆÆ):*:4®4®>"?"I”I”SÇSÇ]€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""ššŠŠ z z"ò"ò$f%f'Ú'Ú)N*N,Â,Â.6/61ª1ª346’6’89;z;z=î=î?b@bBÖBÖDJEJG¾G¾I2J2L¤L¤NOQˆQˆSúSúUlVlXÞXÞZP[P]Â]Â_4`4bgbgdÙdÙfKgKi½i½k/l/nènèpZqZsÌsÌu>v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx7i!1C~!1èC~!1á¾!1dá¾!1›Å!1l›Å!1%Ï!1è%Ï!1[Ò!1´[Ò!1è!1@è!1³("1¼³("1m/"1Äm/"1÷8"1@÷8"1-<"1 -<"1çQ"1˜çQ"1…’"1…’"1?™"1 ?™"1É¢"1˜ É¢"1ÿ¥"1d ÿ¥"1AÓ"1ð AÓ"1£Ú"1Ì £Ú"1ÿ"1H ÿ"1E,#1hE,#1‘:#1D‘:#1Mo#1ÐMo#1�#1ô�#1 š#1  š#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1™€&1à™€&1iR'1$iR'1e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1«ß•a b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [›¬ìö6 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1«ß•a b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [›¬ìö6 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1«ß•a  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [›¬ìö6  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 •µx˜>^ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1[›ö6‘Ѩè ,l  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d[Û6¶‘ìlÇG¢"}ýXØ3³Žéi Ä D   Ÿ  l ì z ú 8 ¸ U Õ  „ 0 °  K¦æx¸A�Ü bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÃÃ)8*84­4­>"?"I•I•ST^@^@hòhòrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD?I�I�ST^u^uhèhèr[s[} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……úúooääY Y"Î"Î$C%C'¸'¸)-*-,¢,¢./1Œ1Œ346v6v8ë8ë:`;`=Õ=Õ?H@HB»B»D.E.G¡G¡IJL‡L‡NúNúPmQmSàSàUSVSXÆXÆZ9[9]¬]¬_`b’b’degxgxiëiëk^l^nÑnÑpDqDs·s·u*v*x�x�z{}ƒ}ƒöö�i‚i„܄܆O‡O‰Á‰Á‹3Œ3ޥޥ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!17~!1Ð7~!1Õ¾!1LÕ¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1 è!1( è!1§("1¤§("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1ÛQ"1€ÛQ"1y’"1üy’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1=Ú"1´ =Ú"1�þ"10 �þ"1ß+#1Pß+#1—7#1,—7#1ã�#1˜ã�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1á€&1Ìá€&1±R'1±R'1Ye'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒ†R b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [›{»ú: bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒ†R b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [›{»ú: bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒ†R  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [›{»ú:  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p�Ìì0 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations12r€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÈHŸvöM Í ²ò €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á áWWÍÍ)C*C4¹4¹>-?-I¡I¡ST^‰^‰hýhýrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±''�� ‰ ‰"ÿ"ÿ$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@…@…BùBùDmEmGáGáIUJULÉLÉN=O=Q±Q±S%T%V™V™X Y [�[�]õ]õ_i`ibÝbÝdQeQgÅgÅi9j9l­l­n!o!q•q•s t v}v}xñxñze{e}Ù}ÙM€M‚Á‚Á„5…5‡©‡©‰ŠŒ�Œ�Ž�‘u‘u“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1³}!1г}!1Q¾!1LQ¾!1›Å!1T›Å!1%Ï!1Ð%Ï!1[Ò!1œ[Ò!1…ç!1(…ç!1#("1¤#("1m/"1¬m/"1÷8"1(÷8"1-<"1ô-<"1WQ"1€WQ"1õ‘"1üõ‘"1?™"1 ?™"1É¢"1€ É¢"1ÿ¥"1L ÿ¥"1AÓ"1Ø AÓ"1aÚ"1´ aÚ"1Áþ"10 Áþ"1,#1P,#1=:#1,=:#1yŽ#1˜yŽ#1™$1$™$1qò$1èqò$1«Þ%1ì«Þ%1!€&1Ì!€&1ñQ'1ñQ'1™d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�Í b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V[›ö6 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�Í b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V[›ö6 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�Í  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V[›ö6  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‰©.N"B bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ù €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1( ¨ ÿ  Ö V ­ - I‰ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×JJ½½)0*04£4£>?I…I…SöSö]g^ghØhØrIsI} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡úúmmààS S"Æ"Æ$9%9'¬'¬)*,’,’./1x1x3ë3ë5^6^8Ñ8Ñ:D;D=·=·?(@(B™B™D E G{G{IìIìK]L]NÎNÎP?Q?S°S°U!V!X’X’Z[]t]t_å_åaVbVdÇdÇf8g8i©i©kln‹n‹püpürmsmuÞuÞwOxOzÀzÀ|1}1¢¢�‚„„„„†õ†õˆe‰e‹Õ‹Õ�EŽE�µ�µ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1=~!1Ð=~!1Û¾!1LÛ¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1è!1(è!1­("1¤­("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1áQ"1€áQ"1’"1ü’"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1aÚ"1´ aÚ"1Áþ"10 Áþ"1,#1P,#1+:#1,+:#1yŽ#1˜yŽ#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1i€&1Ìi€&19R'19R'1ád'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 ÔŠV b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [›{»û; bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 ÔŠV b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [›{»û; bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 ÔŠV  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [›{»û;  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 „¤Ëë4T bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1É €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1$ ¤ û { Ò R © ) I‰ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝQQÅÅ)9*94­4­>?I‘I‘ST^u^uhçhçrYsY} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��……ùùm m"á"á$U%U'É'É)=*=,±,±.%/%1™1™3 4 6�6�8õ8õ:i;i=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b‹b‹dýdýfogoiáiákSlSnÅnÅp7q7s©s©uvx�x�zÿzÿ|q}qãã�U‚U„DŽdž9‡9‰ª‰ª‹ŒŽŒŽŒ�ý�ý’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¹}!1й}!1W¾!1LW¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1‹ç!1(‹ç!1)("1¤)("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1]Q"1€]Q"1û‘"1üû‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1©Ú"1´ ©Ú"1 ÿ"10  ÿ"1K,#1PK,#1—7#1,—7#1³�#1˜³�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ð b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @]�ø8 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ð b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @]�ø8 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ð  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @]�ø8  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~ž,Lr’ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations11q€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¯/„YÙ. ® �Á €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚOOÄÄ)9*94®4®>$?$I˜I˜S T ^€^€hôhôrjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""šš„„ùùn n"ã"ã$X%X'Í'Í)B*B,·,·.,/,1¡1¡346‹6‹89;u;u=ê=ê?_@_BÔBÔDIEIG¾G¾I4J4L¨L¨NOQ�Q�STVxVxXìXìZ`[`]Ô]Ô_H`Hb¼b¼d0e0g¤g¤ijlŒlŒnoqtqtsèsèu\v\xÐxÐzD{D}¸}¸,€,‚ ‚ „…‡ЇЉÿ‰ÿ‹tŒtŽéŽé�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!11~!1Ð1~!1Ͼ!1LϾ!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1è!1(è!1¡("1¤¡("1…/"1¬…/"19"1(9"1E<"1ôE<"1ÕQ"1€ÕQ"1s’"1üs’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"11Ú"1´ 1Ú"1‘þ"10 ‘þ"1Ó+#1PÓ+#1¯7#1,¯7#1¯Ž#1˜¯Ž#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%15�&1Ì5�&1S'1S'1­e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Ç{G b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [›¢âE bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Ç{G b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [›¢âE bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Ç{G  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [›¢âE  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 g‡Ëëþ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Hˆ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  ‚äd Æ F x¸ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg à àYYÒÒ)K*K4Ä4Ä>=?=I´I´S)T)^œ^œhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggààYYÒÒ K K Ä Ä==¶¶//¨¨!!šš Œ Œ"#%~%~'÷'÷)p*p,é,é.b/b1Û1Û3T4T6Í6Í8F9F;¿;¿=8>8@±@±B*C*E£E£GHJ•J•L M OƒOƒQúQúSqTqVèVèX_Y_[Ö[Ö]M^M`Ä`Äb9c9e®e®g#h#j˜j˜l m o€o€qóqósftfvÙvÙxLyL{¿{¿}2~2€¥€¥‚ƒ…‹…‹‡þ‡þ‰pŠpŒâŒâŽT�T‘ƑƓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¹}!1й}!1W¾!1LW¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1‹ç!1(‹ç!1)("1¤)("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1]Q"1€]Q"1û‘"1üû‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1…Ú"1´ …Ú"1åþ"10 åþ"1',#1P',#11:#1,1:#1IŽ#1˜IŽ#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Î b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\œ÷7 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Î b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\œ÷7 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Î  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\œ÷7  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 € Qq0 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Æ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 ƒ Ø X ­ - ‚  V €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØMMÂÂ)7*74¬4¬>!?!I”I”ST^x^xhêhêr^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?M@MBÂBÂD7E7G¬G¬I!J!L”L”NOQzQzSíSíU`V`XÓXÓZF[F]¹]¹_,`,bžbždeg‚g‚iôiôkflfnØnØpJqJs¼s¼u.v.x x z{}„}„öö�h‚h„Ú„Ú†N‡N‰Á‰Á‹4Œ4ާާ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!17~!1Ð7~!1Õ¾!1LÕ¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1 è!1( è!1§("1¤§("1…/"1¬…/"19"1(9"1E<"1ôE<"1ÛQ"1€ÛQ"1y’"1üy’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1OÚ"1´ OÚ"1¯þ"10 ¯þ"1ñ+#1Pñ+#1U:#1,U:#1Q�#1˜Q�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•ÉK b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [›¡áF bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•ÉK b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [›¡áF bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•ÉK  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [›¡áF  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ‡§Ëë> bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ê* €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¨ ( Š l ì N Î Z €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>/?/I¤I¤ST^ˆ^ˆhùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbçbçdZeZgÍgÍi@j@l³l³n$o$q•q•stvwvwxèxèzY{Y}Ê}Ê;€;‚¬‚¬„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1¿}!1п}!1]¾!1L]¾!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1‘ç!1(‘ç!1/("1¤/("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1cQ"1€cQ"1’"1ü’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1eÓ"1Ø eÓ"1»Ú"1´ »Ú"1ÿ"10 ÿ"1],#1P],#1�7#1,�7#1é�#1˜é�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1ù€&1Ìù€&1ÉR'1ÉR'1qe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ñ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Is³N bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ñ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Is³N bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ñ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Is³N  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w—>�¡ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1.n€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¸8‘jêC à ®î €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÇÇ)>*>4µ4µ>.?.I¥I¥ST^Ž^Žhisrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝIVJVLÍLÍNDODQ»Q»S2T2V©V©X Y [—[—]^`…`…bübüdsesgêgêiajalØlØnJoJq¼q¼s.t.v v xy{„{„}ö}öh€h‚Ú‚Ú„L…L‡¾‡¾‰0Š0Œ¢Œ¢Ž�‘†‘†“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1+~!1Ð+~!1ɾ!1Lɾ!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1ýç!1(ýç!1›("1¤›("1y/"1¬y/"19"1(9"19<"1ô9<"1ÏQ"1€ÏQ"1m’"1üm’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1Ú"1´ Ú"1þ"10 þ"1Á+#1PÁ+#1Í7#1,Í7#1u�#1˜u�#1á$1$á$1¹ò$1è¹ò$1óÞ%1ìóÞ%1G�&1ÌG�&1S'1S'1¿e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼p< b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [›{»ù9 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼p< b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [›{»ù9 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼p<  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [›{»ù9  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ëë^~ì  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1^ž€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1xøeåR Ò ? ¿ >~ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßWWÏÏ)G*G4¿4¿>7?7I­I­S#T#^™^™his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ��  "÷"÷$o%o'ç'ç)_*_,×,×.O/O1Ç1Ç3?4?6·6·8/9/;§;§=>@—@—BCE‡E‡GÿGÿIwJwLíLíNcOcQÙQÙSOTOVÅVÅX;Y;[±[±]'^'`�`�bce‰e‰gÿgÿiujulëlënaoaq×q×sMtMvÃvÃx9y9{¯{¯}%~%€›€›‚ƒ…‡…‡‡ý‡ý‰sŠsŒéŒéŽ_�_‘Õ‘Õ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1¿}!1п}!1]¾!1L]¾!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1‘ç!1(‘ç!1/("1¤/("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1cQ"1€cQ"1’"1ü’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1eÓ"1Ø eÓ"1©Ú"1´ ©Ú"1 ÿ"10  ÿ"1K,#1PK,#11:#1,1:#1Ž#1˜Ž#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Ÿ€&1ÌŸ€&1oR'1oR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ï b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jr² M bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ï b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jr² M bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ï  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jr² M  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u•úr’ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ç €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1 œ õ u Î N § ' G‡ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb × ×LLÁÁ)6*64«4«>"?"I—I—S T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbb××LLÁÁ 6 6 « «  ••  ôôiiÞÞS S"È"È$=%='²'²)'*',œ,œ./1†1†3û3û5p6p8å8å:Z;Z=Ï=Ï?D@DB¹B¹D.E.G£G£IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!11~!1Ð1~!1Ͼ!1LϾ!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1ýç!1(ýç!1›("1¤›("1y/"1¬y/"19"1(9"19<"1ô9<"1ÏQ"1€ÏQ"1m’"1üm’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"17Ú"1´ 7Ú"1—þ"10 —þ"1Ù+#1PÙ+#1y:#1,y:#1#�#1˜#�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оt@ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [›{»ú: bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оt@ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [›{»ú: bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оt@  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [›{»ú:  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ëëzš$ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 K €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1, ¬  ™  † ó së+ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÇÇ)=*=4³4³>)?)I�I�ST^…^…hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒuŒuŽéŽé�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1Å}!1ÐÅ}!1c¾!1Lc¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1—ç!1(—ç!15("1¤5("1/"1¬/"1 9"1( 9"1?<"1ô?<"1iQ"1€iQ"1’"1ü’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1»Ú"1´ »Ú"1ÿ"10 ÿ"1],#1P],#1µ7#1,µ7#1¯Ž#1˜¯Ž#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ò b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\œ÷7 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ò b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\œ÷7 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ò  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\œ÷7  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p� *|œ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1D„€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�ôtØX ¼ < t´ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff à àZZÐÐ)F*F4¼4¼>2?2I¦I¦ST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffààZZÔÔ N N È ÈBB¼¼66°°**¤¤ ˜ ˜"#%ˆ%ˆ'þ'þ)t*t,ê,ê.`/`1Ö1Ö3L4L6Â6Â8898;®;®=$>$@š@šBCE†E†GüGüIrJrLæLæNZOZQÎQÎSBTBV¶V¶X*Y*[ž[ž]^`†`†búbúdnengâgâiVjVlÊlÊn>o>q²q²s&t&všvšxy{‚{‚}ö}öj€j‚Þ‚Þ„R…R‡ƇƉ:Š:Œ®Œ®Ž"�"‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1%~!1Ð%~!1þ!1Lþ!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1÷ç!1(÷ç!1•("1¤•("1/"1¬/"1 9"1( 9"1?<"1ô?<"1ÉQ"1€ÉQ"1g’"1üg’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1Ú"1´ Ú"1sþ"10 sþ"1µ+#1Pµ+#1å7#1,å7#1A�#1˜A�#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1�&1Ì�&1áR'1áR'1‰e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±e1 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [››Û÷7 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±e1 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [››Û÷7 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1}±e1  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [››Û÷7  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ËëUuÚú bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1t´€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÐPÈH À @ ¸ 8 D €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Þ ÞUUÌÌ)C*C4º4º>1?1I¦I¦ST^�^�hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIYJYLÎLÎNCOCQ¸Q¸S-T-V¢V¢XY[Œ[Œ]^`v`vbëbëd`e`gÕgÕiJjJl¿l¿n4o4q©q©stv“v“xy{}{}}ò}òg€g‚܂܄Q…Q‡ƇƉ;Š;Œ°Œ°Ž%�%‘š‘š“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1Å}!1ÐÅ}!1c¾!1Lc¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1‘ç!1(‘ç!1/("1¤/("1/"1¬/"1 9"1( 9"1?<"1ô?<"1cQ"1€cQ"1’"1ü’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1¯Ú"1´ ¯Ú"1ÿ"10 ÿ"1Q,#1PQ,#1[:#1,[:#1W�#1˜W�#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ð b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\œ÷7 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ð b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\œ÷7 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ð  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\œ÷7  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 jŠäq‘ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1è( €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  „  h è L Ì X €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÊÊ)>*>4²4²>&?&I˜I˜S T ^|^|hîhîr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$j%j'Þ'Þ)R*R,Æ,Æ.:/:1®1®3"4"6–6–8 9 ;~;~=ò=ò?f@fBÚBÚDNENGÂGÂI6J6L¨L¨NOQŒQŒSþSþUpVpXâXâZT[T]Æ]Æ_8`8bªbªdegŽgŽijlrlrnänäpVqVsÈsÈu:v:x¬x¬z{}�}�€‚t‚t„æ„æ†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1+~!1Ð+~!1ɾ!1Lɾ!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1÷ç!1(÷ç!1•("1¤•("1/"1¬/"1 9"1( 9"1?<"1ô?<"1ÉQ"1€ÉQ"1g’"1üg’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1%Ú"1´ %Ú"1…þ"10 …þ"1Ç+#1PÇ+#1�:#1,�:#1û�#1˜û�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1¥€&1Ì¥€&1uR'1uR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³i5 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [››Û÷7 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³i5 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [››Û÷7 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³i5  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [››Û÷7  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ëëm�ê  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1,l €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1° 0 ¨ (   ˜¼ü €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜRRÇÇ)<*<4±4±>&?&I™I™S T ^^hòhòrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒxxîîd d"Ú"Ú$O%O'Ä'Ä)9*9,®,®.#/#1˜1˜3 4 6‚6‚8÷8÷:l;l=á=á?V@VBËBËD@E@GµGµI*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b¨b¨degŽgŽijltltnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1Ë}!1ÐË}!1i¾!1Li¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1�ç!1(�ç!1;("1¤;("1/"1¬/"1 9"1( 9"1?<"1ô?<"1oQ"1€oQ"1 ’"1ü ’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"17Ú"1´ 7Ú"1—þ"10 —þ"1Ù+#1PÙ+#1Í7#1,Í7#1{�#1˜{�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ó b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\œ÷7 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ó b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\œ÷7 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ó  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\œ÷7  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ëëi‰ø bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Zš€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1hèW×F Æ 5 µ :z €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84­4­> ? I“I“ST^y^yhìhìr_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?L@LB¿B¿D2E2G¥G¥IJL‹L‹NþNþPqQqSäSäUWVWXÊXÊZ=[=]°]°_#`#b–b–d e g|g|iïiïkblbnÕnÕpHqHs»s»u.v.x¡x¡z{}‡}‡úú�m‚m„à„à†S‡S‰ʼnŋ7Œ7ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1~!1Ð~!1½¾!1L½¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1ëç!1(ëç!1‰("1¤‰("1/"1¬/"1 9"1( 9"1?<"1ô?<"1½Q"1€½Q"1[’"1ü[’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1ý7#1,ý7#1 ‘#1˜ ‘#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1ó€&1Ìó€&1ÃR'1ÃR'1ke'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦Z& b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [›}½ö6 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦Z& b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [›}½ö6 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦Z&  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [›}½ö6  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 LlÈ芪 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ŠÊ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1(¨+ « . ® 1 ± Ê  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk å å[[ÑÑ)G*G4½4½>1?1I¥I¥ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkåå__ÙÙ S S Í ÍGGÁÁ;;µµ++¡¡ � �"#%y%y'ï'ï)e*e,Û,Û.Q/Q1Ç1Ç3=4=6³6³8)9);Ÿ;Ÿ=>@‰@‰BýBýDqEqGåGåIYJYLÍLÍNAOAQµQµS)T)V�V�XY[…[…]ù]ù_m`mbábádUeUgÉgÉi=j=l±l±n%o%q™q™s t v�v�xõxõzi{i}Ý}ÝQ€Q‚łń9…9‡­‡­‰ Š Œ“Œ“Ž�‘y‘y“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1Ë}!1ÐË}!1i¾!1Li¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1—ç!1(—ç!15("1¤5("1/"1¬/"1 9"1( 9"1?<"1ô?<"1iQ"1€iQ"1’"1ü’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1 Ú"1´  Ú"1mþ"10 mþ"1¯+#1P¯+#1:#1,:#1)�#1˜)�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%13€&1Ì3€&1R'1R'1«d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ñ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\œ÷7 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ñ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\œ÷7 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ñ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\œ÷7  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ëë_Îî bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 I €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1$ ¤  “  ‚ ñ qé) €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛNNÁÁ)4*44§4§>?I‰I‰SúSú]k^khÜhÜrMsM} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹þþqqääW W"Ê"Ê$=%='°'°)#*#,–,–. / 1|1|3ï3ï5b6b8Õ8Õ:H;H=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éaZbZdËdËf#?#I•I•ST^y^yhëhër]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‰‰ýýq q"å"å$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=á=á?S@SBÅBÅD7E7G©G©IJL�L�NÿNÿPqQqSãSãUUVUXÇXÇZ9[9]«]«_`b�b�degsgsiåiåkWlWnÉnÉp;q;s­s­uvx‘x‘z{}u}uçç�Y‚Y„˄ˆ=‡=‰®‰®‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lOi!1Ñ}!1ÐÑ}!1o¾!1Lo¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1�ç!1(�ç!1;("1¤;("1…/"1¬…/"19"1(9"1E<"1ôE<"1oQ"1€oQ"1 ’"1ü ’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1%Ú"1´ %Ú"1…þ"10 …þ"1Ç+#1PÇ+#1ë7#1,ë7#1G�#1˜G�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1á€&1Ìá€&1±R'1±R'1Ye'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ô b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\œ÷7 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ô b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\œ÷7 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ô  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\œ÷7  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ëëb‚æ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1p°€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1À@º: ´ 4 ® . @ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞSSÈÈ)=*=4²4²>(?(IœIœST^„^„høhørnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžˆˆýýr r"ç"ç$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡އމŠŒxŒxŽíŽí�b‘b“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lOi!1~!1Ð~!1·¾!1L·¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1åç!1(åç!1ƒ("1¤ƒ("1…/"1¬…/"19"1(9"1E<"1ôE<"1·Q"1€·Q"1U’"1üU’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#18#1,8#1Ó‘#1˜Ó‘#1í$1$í$1Åò$1èÅò$1ÿÞ%1ìÿÞ%1G�&1ÌG�&1S'1S'1¿e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›O b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [››Û÷7 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›O b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [››Û÷7 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›O  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [››Û÷7  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Cc¶Ö…¥ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 à€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ Ž  œ  ª * �Ð €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk ä ä]]ÖÖ)O*O4È4È>A?AI¸I¸S-T-^ ^ his†s†} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkää]]ÖÖ O O È ÈAAºº33¬¬%%žž � �" # %‚%‚'û'û)t*t,í,í.f/f1ß1ß3X4X6Ñ6Ñ8J9J;Ã;Ã=<><@µ@µB.C.E§E§G H J™J™LMO‡O‡QþQþSuTuVìVìXcYc[Ú[Ú]Q^Q`È`Èb=c=e²e²g'h'jœjœlmo„o„q÷q÷sjtjvÝvÝxPyP{Ã{Ã}6~6€©€©‚ƒ…�…�‡ˆŠtŠtŒæŒæŽX�X‘ʑʓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lOi!1Ë}!1ÐË}!1i¾!1Li¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1�ç!1(�ç!1;("1¤;("1…/"1¬…/"19"1(9"1E<"1ôE<"1oQ"1€oQ"1 ’"1ü ’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1ûÙ"1´ ûÙ"1[þ"10 [þ"1�+#1P�+#1£:#1,£:#1‘#1˜‘#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ò b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @\œ÷7 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ò b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @\œ÷7 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ò  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @\œ÷7  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ËëTt¸Ø bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1*j €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¨ ( ¢ " œ –ºú €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜQQÆÆ);*;4°4°>%?%I˜I˜S T ^|^|hîhîrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒvvëë` `"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lOi!1~!1Ð~!1·¾!1L·¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1ëç!1(ëç!1‰("1¤‰("1…/"1¬…/"19"1(9"1E<"1ôE<"1½Q"1€½Q"1[’"1ü[’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1å:#1,å:#1¥’#1˜¥’#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1Õ€&1ÌÕ€&1¥R'1¥R'1Me'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�S b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 [››Û÷7 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�S b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 [››Û÷7 bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�S  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 [››Û÷7  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ss¶Ö†¦ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1n® €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¸ 8 Æ FÔTâb^ž €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii à àWWÎÎ)E*E4¼4¼>3?3I¨I¨ST^Œ^Œhýhýroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiààWWÎÎ E E ¼ ¼33ªª!!˜˜††ýýt t"ë"ë$b%b'Ù'Ù)P*P,Ç,Ç.>/>1µ1µ3,4,6£6£89;‘;‘=>@@BöBöDmEmGäGäI[J[LÐLÐNEOEQºQºS/T/V¤V¤XY[Ž[Ž]^`x`xbëbëd^e^gÑgÑiDjDl·l·n(o(q™q™s t v{v{xìxìz]{]}Î}Î?€?‚°‚°„!…!‡“‡“‰ŠŒuŒu޿޿�W‘W“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1é}!1�é}!1‡¾!1‡¾!1¿Â!1$¿Â!1»ç!1�»ç!1Y("1Y("1‘,"1$‘,"1�Q"1��Q"1+’"1+’"1c–"1$c–"1Ù"1�Ù"1wý"1üwý"1;0#1 ;0#1#1| #1ó$1Ü ó$1Ëò$1  Ëò$1ß%1¤ ß%1�&1„ �&1íR'1ÈíR'1•e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14s§æÏ› b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14s§æÏ› b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14s§æÏ› b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ëë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww î îeeÜÜ)S*S4Ê4Ê>A?AI¸I¸S/T/^¡^¡his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwîîeeÜÜ S S Ê ÊAA¸¸//¦¦””  ‚ ‚"ù"ù$p%p'ç'ç)^*^,Õ,Õ.L/L1Ã1Ã3:4:6±6±8(9(;Ÿ;Ÿ=>@�@�BCE{E{GòGòIiJiLàLàNWOWQÎQÎSETEV¼V¼X3Y3[ª[ª]!^!`˜`˜bce†e†gýgýitjtlëlën]o]qÏqÏsAtAv³v³x%y%{—{—} ~ €{€{‚í‚í„_…_‡чщCŠCŒµŒµŽ'�'‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1~!1�~!1±¾!1 ±¾!1SÅ!1SÅ!1åç!1€åç!1ƒ("1üƒ("1%/"1%/"1·Q"1p·Q"1U’"1ìU’"1÷˜"1ô÷˜"1ÁÚ"1`ÁÚ"1!ÿ"1Ì!ÿ"1;0#1ì;0#1#1L #1ÿ$1¬ ÿ$1×ò$1p ×ò$1ß%1t ß%1e�&1T e�&15S'1˜5S'1Ýe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�C b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1“Ó€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�C b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1“Ó€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�C b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1“Ó €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1q‘  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ì ìddÜÜ)T*T4Ì4Ì>B?BI¸I¸S.T.^¤^¤his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttììddÜÜ T T Ì ÌDD¼¼44¬¬$$œœ Œ Œ"#%|%|'ô'ô)l*l,ä,ä.\/\1Ô1Ô3L4L6Ä6Ä8<9<;´;´=,>,@¤@¤BCE”E”G H J‚J‚LøLøNnOnQäQäSZTZVÐVÐXFYF[¼[¼]2^2`¨`¨bce”e”g h j€j€lölönlolqâqâsXtXvÎvÎxDyD{º{º}0~0€¦€¦‚ƒ…’…’‡ˆŠ~Š~ŒôŒôŽj�j‘à‘à“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1ã}!1�ã}!1�¾!1�¾!1¿Â!1$¿Â!1µç!1�µç!1S("1S("1‘,"1$‘,"1‡Q"1�‡Q"1%’"1%’"1c–"1$c–"1Ù"1�Ù"1wý"1üwý"1;0#1 ;0#1#1| #1á$1Ü á$1¹ò$1  ¹ò$1óÞ%1¤ óÞ%1À&1„ À&1“R'1È“R'1;e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14q¥âМ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14q¥âМ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14q¥âМ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ëë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu ê ê__ÔÔ)I*I4¾4¾>3?3I¨I¨ST^�^�hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuêê__ÔÔ I I ¾ ¾33¨¨’’||ññf f"Û"Û$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I+J+L L NOQŠQŠSÿSÿUtVtXéXéZ^[^]Ó]Ó_H`Hb½b½d2e2g§g§ijl‘l‘noququsçsçuYvYxËxËz={=}¯}¯!€!‚“‚“„…‡w‡w‰é‰é‹[Œ[ŽÍŽÍ�?‘?“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1~!1�~!1±¾!1 ±¾!1SÅ!1SÅ!1åç!1€åç!1ƒ("1üƒ("1%/"1%/"1·Q"1p·Q"1U’"1ìU’"1÷˜"1ô÷˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1;0#1ì;0#1#1L #1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1í€&1T í€&1½R'1˜½R'1ee'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�E b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1’Ò€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�E b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1’Ò€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1[�E b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1’Ò €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1|œ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ê ê``ÖÖ)L*L4Â4Â>6?6IªIªST^’^’hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttêê``ÖÖ L L  Â88®®$$šš††üür r"è"è$^%^'Ô'Ô)J*J,À,À.6/61¬1¬3"4"6˜6˜89;„;„=ú=ú?p@pBæBæD\E\GÒGÒIFJFLºLºN.O.Q¢Q¢STVŠVŠXþXþZr[r]æ]æ_Z`ZbÎbÎdBeBg¶g¶i*j*lžlžnoq†q†súsúunvnxâxâzV{V}Ê}Ê>€>‚²‚²„&…&‡š‡š‰ŠŒ‚Œ‚ŽöŽö�j‘j“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1ã}!1�ã}!1�¾!1 �¾!1¹Â!1¹Â!1µç!1€µç!1S("1üS("1‹,"1‹,"1‡Q"1p‡Q"1%’"1ì%’"1]–"1ô]–"1Ù"1`Ù"1wý"1Ìwý"150#1ì50#1#1L #1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1)�&1T )�&1ùR'1˜ùR'1¡e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤ã b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤ã b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤ã b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ëë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww ñ ñkkáá)W*W4Í4Í>A?AIµIµS)T)^�^�his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwññkkåå _ _ Ù ÙSSÍÍGGÁÁ;;µµ// © ©"###%™%™'(*…*…,û,û.q/q1ç1ç3]4]6Ó6Ó8I9I;¿;¿=5>5@«@«B!C!E—E—G H J�J�LõLõNiOiQÝQÝSQTQVÅVÅX9Y9[­[­]!^!`•`•b c e}e}gñgñiejelÙlÙnMoMqÁqÁs5t5v©v©xy{‘{‘}~€y€y‚í‚í„a…a‡Õ‡Õ‰IŠIŒ½Œ½Ž1�1‘¥‘¥“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1·¾!1 ·¾!1SÅ!1SÅ!1åç!1€åç!1ƒ("1üƒ("1%/"1%/"1·Q"1p·Q"1U’"1ìU’"1÷˜"1ô÷˜"1µÚ"1`µÚ"1ÿ"1Ìÿ"150#1ì50#1#1L #1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1#�&1T #�&1óR'1˜óR'1›e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜L b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1›Û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜L b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1›Û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜L b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1›Û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1lŒ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp ç ç^^ÕÕ)L*L4Ã4Ã>8?8I­I­S"T"^—^—h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppçç^^ÕÕ L L à Ã::±±((ŸŸ�� { {"ò"ò$i%i'à'à)W*W,Î,Î.E/E1¼1¼33436ª6ª8!9!;˜;˜=>@†@†BýBýDtEtGëGëI`J`LÕLÕNJOJQ¿Q¿S4T4V©V©XY[“[“]^`}`}bòbòdgeggÜgÜiQjQlÆlÆn;o;q°q°s%t%všvšxy{„{„}ù}ùn€n‚ã‚ã„X…X‡͇͉BŠBŒ·Œ·Ž,�,‘¡‘¡“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Ý}!1�Ý}!1{¾!1 {¾!1¹Â!1¹Â!1¯ç!1€¯ç!1M("1üM("1‹,"1‹,"1�Q"1p�Q"1’"1ì’"1]–"1ô]–"1Ù"1`Ù"1wý"1Ìwý"150#1ì50#1#1L #1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1±€&1T ±€&1�R'1˜�R'1)e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢ß b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢ß b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢ß b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ëë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu í íeeÙÙ)M*M4Á4Á>3?3I¥I¥ST^‰^‰hûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuííeeÝÝ U U Í ÍEE½½55­­%%�� � �"#%y%y'í'í)a*a,Õ,Õ.I/I1½1½31416¥6¥89;�;�=>@u@uBéBéD]E]GÑGÑICJCLµLµN'O'Q™Q™S T V}V}XïXïZa[a]Ó]Ó_E`Eb·b·d)e)g›g›i j llnñnñpcqcsÕsÕuGvGx¹x¹z+{+}�}�€‚�‚�„ó„ó†e‡e‰׉׋IŒI޻޻�-‘-“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1·¾!1 ·¾!1SÅ!1SÅ!1ëç!1€ëç!1‰("1ü‰("1%/"1%/"1½Q"1p½Q"1[’"1ì[’"1÷˜"1ô÷˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"150#1ì50#1#1L #1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1½€&1T ½€&1�R'1˜�R'15e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜N b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1›Û€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜N b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1›Û€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜N b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1›Û €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1{›  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ç ç]]ÒÒ)G*G4¼4¼>/?/I¢I¢ST^ˆ^ˆhûhûrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqçç]]ÓÓ I I ¿ ¿55««!!——  ƒƒùùo o"å"å$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ì=ì?a@aBÖBÖDKEKGÀGÀI3J3L¦L¦NOQŒQŒSÿSÿUrVrXåXåZX[X]Ë]Ë_>`>b±b±d$e$g—g—i j l}l}nðnðpcqcsÖsÖuIvIx¼x¼z/{/}¢}¢€‚ˆ‚ˆ„û„û†n‡n‰á‰á‹TŒTŽÇŽÇ�:‘:“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Ý}!1�Ý}!1{¾!1 {¾!1¿Â!1¿Â!1©ç!1€©ç!1G("1üG("1‘,"1‘,"1{Q"1p{Q"1’"1ì’"1c–"1ôc–"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"150#1ì50#1#1L #1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡à b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡à b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡à b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1f†  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æ[[ÐÐ)E*E4¸4¸>+?+IžIžST^„^„h÷h÷rhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  €€õõj j"ß"ß$T%T'É'É)>*>,³,³.(/(1�1�346‡6‡8ü8ü:q;q=ä=ä?W@WBÊBÊD=E=G°G°I#J#L–L–N O Q|Q|SïSïUbVbXÕXÕZH[H]»]»_.`.b¡b¡deg‡g‡iúiúkmlmnànàpSqSsÆsÆu9v9x¬x¬z{}’}’€‚x‚x„ë„ë†\‡\‰Ή΋@Œ@޲޲�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1½¾!1 ½¾!1/Å!1/Å!1ëç!1€ëç!1‰("1ü‰("1/"1/"1½Q"1p½Q"1[’"1ì[’"1Ó˜"1ôÓ˜"1Ù"1`Ù"1qý"1Ìqý"150#1ì50#1#1L #1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1ù€&1T ù€&1ÉR'1˜ÉR'1qe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡U! b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1{»€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡U! b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1{»€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡U! b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1{» €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ëë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp ê ê``ÖÖ)L*L4À4À>4?4I¨I¨ST^�^�hisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppêêddÞÞ X X Ò ÒLLÆÆ@@ºº00¦¦ ’ ’"#%~%~'ô'ô)j*j,à,à.V/V1Ì1Ì3B4B6¸6¸8.9.;¤;¤=>@Œ@ŒBCEtEtGèGèI\J\LÐLÐNDODQ¸Q¸S,T,V V XY[ˆ[ˆ]ü]ü_p`pbäbädXeXgÌgÌi@j@l´l´n(o(qœqœstv„v„xøxøzl{l}à}àT€T‚ȂȄ<…<‡®‡®‰!Š!Œ”Œ”Ž�‘z‘z“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1×}!1�×}!1u¾!1 u¾!1¿Â!1¿Â!1©ç!1€©ç!1G("1üG("1‘,"1‘,"1{Q"1p{Q"1’"1ì’"1c–"1ôc–"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"150#1ì50#1#1L #1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1?€&1T ?€&1R'1˜R'1·d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸÜ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸÜ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸÜ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1k‹  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æYYÌÌ)?*?4°4°>!?!I’I’ST^t^thåhårTsT} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  ||ïïb b"Õ"Õ$H%H'»'»).*.,¡,¡./1‡1‡3ú3ú5m6m8à8à:S;S=Ä=Ä?5@5B¦B¦DEGˆGˆIùIùKjLjNÛNÛPLQLS½S½U.V.XŸXŸZ[]�]�_ò_òacbcdÔdÔfEgEi¶i¶k'l'n˜n˜p q szszuëuëw\x\zÍzÍ|>}>¯¯� ‚ „‘„‘†‡‰p‰p‹à‹à�PŽP�À�À’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1½¾!1 ½¾!1/Å!1/Å!1ñç!1€ñç!1�("1ü�("1/"1/"1ÃQ"1pÃQ"1a’"1ìa’"1Ó˜"1ôÓ˜"1Ù"1`Ù"1qý"1Ìqý"150#1ì50#1#1L #1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1�€&1T �€&1QR'1˜QR'1ùd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡W# b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1{»€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡W# b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1{»€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡W# b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1{» €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ëë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp è è\\ÐÐ)D*D4¶4¶>(?(IšIšS T ^~^~hðhðr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppèè``ØØ P P È È@@¸¸00¨¨�� x x"ì"ì$`%`'Ô'Ô)H*H,¼,¼.0/01¤1¤346Œ6Œ89;t;t=æ=æ?X@XBÊBÊD[>]°]°_"`"b”b”degxgxiêiêk\l\nÎnÎp@q@s²s²u$v$x–x–z{}z}zìì�^‚^„ЄІ@‡@‰±‰±‹"Œ"ޓޓ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1×}!1�×}!1u¾!1 u¾!1³Â!1³Â!1£ç!1€£ç!1A("1üA("1…,"1…,"1uQ"1puQ"1’"1ì’"1W–"1ôW–"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1á€&1T á€&1±R'1˜±R'1Ye'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžÝ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžÝ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžÝ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1k‹  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo ç ç\\ÑÑ)F*F4»4»>/?/I£I£ST^‹^‹hÿhÿrsss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooçç__×× O O Ç Ç??··//§§‘‘ { {"ð"ð$e%e'Ú'Ú)O*O,Ä,Ä.9/91®1®3#4#6˜6˜8 9 ;‚;‚=÷=÷?l@lBáBáDVEVGËGËI?J?L³L³N'O'Q›Q›STVƒVƒX÷X÷Zk[k]ß]ß_S`SbÇbÇd;e;g¯g¯i#j#l—l—n o qqsósóugvgxÛxÛzO{O}Ã}Ã7€7‚«‚«„…‡“‡“‰ŠŒ}Œ}ŽòŽò�g‘g“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1~!1�~!1½¾!1 ½¾!1YÅ!1YÅ!1ñç!1€ñç!1�("1ü�("1+/"1+/"1ÃQ"1pÃQ"1a’"1ìa’"1ý˜"1ôý˜"1 Ù"1` Ù"1ký"1Ìký"1/0#1ì/0#1û~#1L û~#1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1A�&1T A�&1S'1˜S'1¹e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vª^* b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¤ä€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vª^* b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¤ä€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vª^* b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¤ä €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ëë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll å å^^××)P*P4É4É>@?@I·I·S,T,^Ÿ^Ÿhisƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllåå^^×× P P É ÉBB»»44­­&&ŸŸ ‘ ‘" # %ƒ%ƒ'ü'ü)u*u,î,î.g/g1à1à3Y4Y6Ò6Ò8K9K;Ä;Ä==>=@¶@¶B/C/E¨E¨G!H!J˜J˜LMO†O†QýQýStTtVëVëXbYb[Ù[Ù]P^P`Ç`Çb*?*I�I�ST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““}}òòg g"Ü"Ü$Q%Q'Æ'Æ);*;,°,°.%/%1š1š346„6„8ù8ù:n;n=ã=ã?X@XBÍBÍDBEBG·G·I*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b§b§deg‹g‹iýiýkolonánápSqSsÅsÅu7v7x©x©z{}�}�ÿÿ�q‚q„ã„ã†U‡U‰ȉȋ;Œ;ޮޮ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1%~!1�%~!1þ!1 þ!1YÅ!1YÅ!1ñç!1€ñç!1�("1ü�("1+/"1+/"1ÃQ"1pÃQ"1a’"1ìa’"1ý˜"1ôý˜"1 Ù"1` Ù"1ký"1Ìký"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Ï€&1T Ï€&1ŸR'1˜ŸR'1Ge'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vª`, b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¥å€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vª`, b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¥å€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vª`, b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¥å €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ëë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ã ãZZÑÑ)H*H4¿4¿>4?4I©I©ST^�^�hþhþrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllããZZÑÑ H H ¿ ¿66­­$$››‰‰ w w"î"î$e%e'Ü'Ü)S*S,Ê,Ê.A/A1¸1¸3/4/6¦6¦89;”;”= > @‚@‚BùBùDpEpGçGçI\J\LÑLÑNFOFQ»Q»S0T0V¥V¥XY[�[�]^`y`ybìbìd_e_gÒgÒiEjEl¸l¸n)o)qšqšs t v|v|xíxíz^{^}Ï}Ï@€@‚±‚±„"…"‡’‡’‰ŠŒtŒtŽåŽå�V‘V“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ë}!1�Ë}!1i¾!1 i¾!1¿Â!1¿Â!1�ç!1€�ç!1;("1ü;("1‘,"1‘,"1oQ"1poQ"1 ’"1ì ’"1c–"1ôc–"1£Ú"1`£Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›Ú b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›Ú b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›Ú b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1hˆ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk â âYYÐÐ)G*G4¾4¾>5?5I¬I¬S#T#^•^•hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkââYYÐÐ G G ¾ ¾55¬¬##ššˆˆÿÿv v"í"í$d%d'Û'Û)R*R,É,É.@/@1·1·3.4.6¥6¥89;“;“= > @�@�BøBøDoEoGæGæI]J]LÔLÔNKOKQÂQÂS9T9V°V°X'Y'[ž[ž]^`Œ`ŒbcezezgñgñihjhlßlßnQoQqÃqÃs5t5v§v§xy{‹{‹}ý}ýo€o‚á‚á„S…S‡Ňʼn7Š7Œ©Œ©Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1%~!1�%~!1þ!1 þ!1)Å!1)Å!1÷ç!1€÷ç!1•("1ü•("1û."1û."1ÉQ"1pÉQ"1g’"1ìg’"1͘"1ô͘"1 Ù"1` Ù"1ký"1Ìký"1)0#1ì)0#1õ~#1L õ~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1G�&1T G�&1S'1˜S'1¿e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³g3 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1|¼€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³g3 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1|¼€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³g3 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1|¼ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ëë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àXXÐÐ)H*H4À4À>6?6I¬I¬S"T"^˜^˜his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  �� € €"ø"ø$p%p'è'è)`*`,Ø,Ø.P/P1È1È3@4@6¸6¸8090;¨;¨= > @˜@˜BCEˆEˆGHJvJvLìLìNbObQØQØSNTNVÄVÄX:Y:[°[°]&^&`œ`œbceˆeˆgþgþitjtlêlên`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ë}!1�Ë}!1i¾!1 i¾!1¿Â!1¿Â!1—ç!1€—ç!15("1ü5("1‘,"1‘,"1iQ"1piQ"1’"1ì’"1c–"1ôc–"1ÁÚ"1`ÁÚ"1!ÿ"1Ì!ÿ"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™ Ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™ Ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™ Ö b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ƒ£  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii Þ ÞSSÈÈ)=*=4²4²>'?'IœIœST^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiÞÞSSÈÈ = = ² ²''œœ††ûûppååZ Z"Ï"Ï$D%D'¹'¹).*.,£,£./1�1�346w6w8ì8ì:a;a=Ö=Ö?K@KBÀBÀD5E5GªGªIJL”L”N O Q~Q~SóSóUhVhXÝXÝZR[R]Ç]Ç_<`*?*IžIžST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽzzððf f"Ü"Ü$R%R'È'È)>*>,´,´.*/*1 1 346Œ6Œ89;x;x=î=î?d@dBÚBÚDPEPGÆGÆI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Å}!1�Å}!1c¾!1 c¾!1­Â!1­Â!1—ç!1€—ç!15("1ü5("1,"1,"1iQ"1piQ"1’"1ì’"1Q–"1ôQ–"1�Ú"1`�Ú"1ýþ"1Ìýþ"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1 �&1T  �&1ÛR'1˜ÛR'1ƒe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜ × b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜ × b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜ × b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1a�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk å å__ÕÕ)K*K4Á4Á>5?5I©I©ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkåå__ÙÙ S S Í ÍGGÁÁ;;µµ//©©## � �"#%�%�'(*y*y,ï,ï.e/e1Û1Û3Q4Q6Ç6Ç8=9=;³;³=)>)@Ÿ@ŸBCE‹E‹GHJuJuLéLéN]O]QÑQÑSETEV¹V¹X-Y-[¡[¡]^`‰`‰býbýdqeqgågåiYjYlÍlÍnAoAqµqµs)t)v�v�xy{…{…}ù}ùm€m‚á‚á„U…U‡ɇɉ=Š=Œ±Œ±Ž%�%‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1+~!1�+~!1ɾ!1 ɾ!1MÅ!1MÅ!1ýç!1€ýç!1›("1ü›("1/"1/"1ÏQ"1pÏQ"1m’"1ìm’"1ñ˜"1ôñ˜"1Ù"1`Ù"1eý"1Ìeý"1)0#1ì)0#1õ~#1L õ~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼p< b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 à€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼p< b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 à€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼p< b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 à €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ëë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>,?,I¡I¡ST^‹^‹hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbæbæd[e[gÐgÐiEjElºlºn/o/q¤q¤stvŽvŽxy{x{x}í}íb€b‚ׂׄL…L‡Á‡Á‰6Š6Œ«Œ«Ž � ‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Å}!1�Å}!1c¾!1 c¾!1­Â!1­Â!1‘ç!1€‘ç!1/("1ü/("1,"1,"1cQ"1pcQ"1’"1ì’"1Q–"1ôQ–"1ÁÚ"1`ÁÚ"1!ÿ"1Ì!ÿ"1)0#1ì)0#1õ~#1L õ~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–Ó b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–Ó b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–Ó b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ˆ¨  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áYYÍÍ)A*A4µ4µ>'?'I™I™S T ^}^}hïhïrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‘‘  � �"ù"ù$m%m'á'á)U*U,É,É.=/=1±1±3%4%6™6™8 9 ;�;�=õ=õ?i@iBÝBÝDQEQGÅGÅI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1+~!1�+~!1ɾ!1 ɾ!1MÅ!1MÅ!1ýç!1€ýç!1›("1ü›("1/"1/"1ÏQ"1pÏQ"1m’"1ìm’"1ñ˜"1ôñ˜"1Ù"1`Ù"1eý"1Ìeý"1)0#1ì)0#1õ~#1L õ~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼r> b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¡á€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼r> b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¡á€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ˆ¼r> b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¡á €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ëë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÆÆ);*;4°4°>#?#I–I–S T ^|^|hïhïrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnänäpWqWsÊsÊu=v=x°x°z#{#}–}– € ‚|‚|„ï„ï†b‡b‰Õ‰Õ‹HŒH޻޻�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¿}!1�¿}!1]¾!1 ]¾!1³Â!1³Â!1‘ç!1€‘ç!1/("1ü/("1…,"1…,"1cQ"1pcQ"1’"1ì’"1W–"1ôW–"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ô b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ô b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ô b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ëë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94¬4¬>?I’I’ST^z^zhíhír^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ø=Ø?K@KB¾B¾D1E1G¤G¤IJLŠLŠNýNýPpQpSãSãUVVVXÉXÉZ<[<]¯]¯_"`"b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†R‡R‰ĉċ6Œ6ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!11~!1�1~!1Ͼ!1 Ͼ!1#Å!1#Å!1è!1€è!1¡("1ü¡("1õ."1õ."1ÕQ"1pÕQ"1s’"1ìs’"1ǘ"1ôǘ"1…Ú"1`…Ú"1åþ"1Ìåþ"1#0#1ì#0#1ï~#1L ï~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ç€&1T ç€&1·R'1˜·R'1_e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘ÅyE b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1{»€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘ÅyE b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1{»€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘ÅyE b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1{» €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Mm  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Þ ÞTTÊÊ)@*@4´4´>(?(IœIœST^Ž^Žhistst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÞÞXXÒÒ L L Æ Æ@@ºº44®®$$šš † †"ü"ü$r%r'è'è)^*^,Ô,Ô.J/J1À1À36466¬6¬8"9";˜;˜= > @€@€BôBôDhEhGÜGÜIPJPLÄLÄN8O8Q¬Q¬S T V”V”XY[|[|]ð]ð_d`dbâbâdVeVgÊgÊi>j>l²l²n&o&qšqšstv‚v‚xöxözj{j}Þ}ÞR€R‚ƂƄ:…:‡¬‡¬‰ŠŒ’Œ’Ž�‘x‘x“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¹}!1�¹}!1W¾!1 W¾!1³Â!1³Â!1‹ç!1€‹ç!1)("1ü)("1…,"1…,"1]Q"1p]Q"1û‘"1ìû‘"1W–"1ôW–"1ÿØ"1`ÿØ"1_ý"1Ì_ý"1#0#1ì#0#1ï~#1L ï~#1¥$1¬ ¥$1}ò$1p }ò$1·Þ%1t ·Þ%1'€&1T '€&1÷Q'1˜÷Q'1Ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ð b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ð b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ð b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ëë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚMMÀÀ)3*34¤4¤>?I†I†SûSû]l^lhÝhÝrLsL} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠýýppããV V"É"É$<%<'¯'¯)"*",•,•./1{1{3î3î5a6a8Ô8Ô:G;G=¸=¸?)@)BšBšD E G|G|IíIíK^L^NÏNÏP@Q@S±S±U"V"X“X“Z[]u]u_æ_æa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!11~!1�1~!1Ͼ!1 Ͼ!1#Å!1#Å!1è!1€è!1¡("1ü¡("1õ."1õ."1ÕQ"1pÕQ"1s’"1ìs’"1ǘ"1ôǘ"1�Ú"1`�Ú"1ýþ"1Ìýþ"1#0#1ì#0#1ï~#1L ï~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1o€&1T o€&1?R'1˜?R'1çd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘Å{G b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1{»€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘Å{G b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1{»€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‘Å{G b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1{» €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1jŠ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜPPÄÄ)8*84ª4ª>?IŽIŽST^z^zhìhìr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœ„„øøl l"à"à$T%T'È'È)<*<,°,°.$/$1˜1˜3 4 6€6€8ô8ô:h;h=Ú=Ú?L@LB¾B¾D0E0G¢G¢IJL†L†NøNøPjQjSÜSÜUNVNXÀXÀZ2[2]¤]¤_`b�b�degtgtiæiækXlXnÊnÊp%?%I™I™ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››……úúo o"ä"ä$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`IbÉbÉd=e=g±g±i%j%l™l™n o q�q�sõsõuivixÝxÝzQ{Q}Å}Å9€9‚­‚­„!…!‡•‡•‰ Š ŒŒŽôŽô�i‘i“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!17~!1�7~!1Õ¾!1 Õ¾!1AÅ!1AÅ!1è!1€è!1¡("1ü¡("1/"1/"1ÕQ"1pÕQ"1s’"1ìs’"1å˜"1ôå˜"1ÛØ"1`ÛØ"1;ý"1Ì;ý"10#1ì0#1é~#1L é~#1á$1¬ á$1 ñ$1p  ñ$1çÀ%1t çÀ%1;�&1T ;�&11Q'1˜1Q'1Ùc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎ‚N b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1žÞ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎ‚N b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1žÞ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎ‚N b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1žÞ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ËÇ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ù ÙRRÕÕ)N*N4Ç4Ç>>?>IµIµS*T*^©^©his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÙÙRRËË D D ½ ½66¯¯((¡¡““  … …"þ"þ$�%�'ú'ú)s*s,ì,ì.e/e1Þ1Þ3W4W6Ð6Ð8I9I;Â;Â=;>;@´@´B-C-E¦E¦GHJ–J–L M O„O„QûQûSrTrVéVéX`Y`[×[×]N^N`Å`Åb:c:e¯e¯g$h$j™j™lmo�o�qrtstsvævæxYyY{Ì{Ì}?~?€²€²‚%ƒ%…˜…˜‡ ˆ Š{Š{ŒíŒíŽ_�_‘ёѓ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!1³}!1�³}!1Q¾!1 Q¾!1¡Â!1¡Â!1…ç!1€…ç!1#("1ü#("1s,"1s,"1WQ"1pWQ"1õ‘"1ìõ‘"1E–"1ôE–"1CÚ"1`CÚ"1£þ"1Ô£þ"10#1ô0#1é~#1T é~#1½$1´ ½$1åð$1x åð$1ÃÀ%1| ÃÀ%1�€&1\ �€&1ƒP'1 ƒP'1+c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�Í b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�Í b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1\�Í b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1ØhÃ/  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×LLÅÅ):*:4¯4¯>"?"I•I•ST^…^…h÷h÷risi} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡üüqqææ[ ["Ð"Ð$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I"J"L•L•NOQ{Q{SîSîUaVaXÔXÔZG[G]º]º_-`-b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰̉̋?Œ?޲޲�%‘%“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!17~!1�7~!1Õ¾!1 Õ¾!1AÅ!1AÅ!1 è!1€ è!1§("1ü§("1/"1/"1ÛQ"1pÛQ"1y’"1ìy’"1å˜"1ôå˜"1ÛØ"1`ÛØ"1;ý"1Ì;ý"10#1ì0#1é~#1L é~#1É$1¬ É$1ñð$1p ñð$1ÏÀ%1t ÏÀ%1É€&1T É€&1¿P'1˜¿P'1gc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎ„P b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ÿß€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎ„P b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ÿß€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1šÎ„P b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ÿß €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ËÇ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` × ×NNÍÍ)D*D4»4»>0?0I¥I¥ST^•^•hisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``××NNÅÅ < < ³ ³**¡¡��}}ôôk k"â"â$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n1o1q¢q¢stv„v„xõxõzf{f}×}×H€H‚¹‚¹„*…*‡š‡š‰ Š Œ|Œ|ŽíŽí�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„1i!1k}!1k}!1 ¾!1| ¾!1mÃ!1„mÃ!1£Æ!1ð£Æ!1-Ð!1|-Ð!1cÓ!1HcÓ!1=ç!1Ô=ç!1Û'"1PÛ'"1?-"1X?-"1u0"1Äu0"1ÿ9"1Pÿ9"15="15="1Q"1¨Q"1­‘"1$ ­‘"1—"1, —"1Gš"1˜ Gš"1Ñ£"1$ Ñ£"1§"1ð §"1IÔ"1| IÔ"1Ú"1XÚ"1sþ"1Ôsþ"1µ+#1ôµ+#1—:#1З:#1So#1\So#13�#1€3�#11š#1¬1š#1Û$18Û$1³ò$1ü³ò$1íÞ%1íÞ%1gì%1àgì%15�&1l5�&1S'1°S'1­e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PÅ‘ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Í  bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PÅ‘ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Í   bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PÅ‘  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Í  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ííe…ë  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1F]�ú:—×4t  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d†]ÝÑQ:ºœ—gçôtÑQ®.‹ hèEÅ"¢ÿ Ü \ ¹ 9 –  s ó ¶ ö P Ð -mÊ g§D bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ù ÙÕÕ)N*N4Ç4Ç>@?@I·I·S,T,^«^«his‘s‘} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÙÙRRËË D D ½ ½66¯¯((¡¡ÚÚSSÌÌE E"¾"¾$�%�'ú'ú)s*s,ì,ì.e/e1Þ1Þ3W4W6Ð6Ð8I9I;Â;Â=;>;@´@´B-C-E¦E¦GHJ˜J˜LMO†O†QýQýStTtVëVëXbYb[Ù[Ù]P^P`Ç`Çb3?3I§I§ST^›^›hisƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯))££ — —"#%I%I'¿'¿)5*5,«,«.!/!1×1×3M4M6Ã6Ã8999;¯;¯=%>%@›@›BCE‡E‡GýGýIsJsLçLçN[O[QÏQÏSCTCV·V·X+Y+[Ÿ[Ÿ]^`‡`‡bûbûdoeogãgãiWjWlËlËnKoKq¿q¿s3t3v§v§xy{�{�}~€w€w‚ë‚ë„_…_‡Ó‡Ó‰GŠGŒ»Œ»Ž/�/‘£‘£“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„1i!1k}!1k}!1 ¾!1| ¾!1mÃ!1„mÃ!1£Æ!1ð£Æ!1-Ð!1|-Ð!1cÓ!1HcÓ!17ç!1Ô7ç!1Õ'"1PÕ'"1?-"1X?-"1u0"1Äu0"1ÿ9"1Pÿ9"15="15="1 Q"1¨ Q"1§‘"1$ §‘"1—"1, —"1Gš"1˜ Gš"1Ñ£"1$ Ñ£"1§"1ð §"1IÔ"1| IÔ"1CÚ"1XCÚ"1£þ"1Ô£þ"1å+#1ôå+#1Í:#1ÐÍ:#1‰o#1\‰o#15�#1€5�#13›#1¬3›#1Ã$18Ã$1›ò$1ü›ò$1ÕÞ%1ÕÞ%1Oì%1àOì%1À&1lÀ&1“R'1°“R'1;e'1´€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OÂŽ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Í  bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OÂŽ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Í   bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OÂŽ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Í  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Íív–< bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1k«‘ÑV¿ÿh¨  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dkëVÖAÁ,¬2²—ý}‚ÈHím“ØXÁAª* “  | ü e å N Î 7 · A �    K¶ö_ŸH bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` × ×  ÍÍ)D*D4»4»>2?2I§I§ST^—^—hiszsz} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``××NNÅÅ < < ³ ³**¡¡��ÅÅ<<³³* *"¡"¡$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbêbêd]e]gÐgÐiCjCl¶l¶n3o3q¤q¤stv†v†x÷x÷zh{h}Ù}ÙJ€J‚»‚»„,…,‡ž‡ž‰ŠŒ€Œ€ŽñŽñ�b‘b“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1O~!1èO~!1í¾!1dí¾!1›Å!1l›Å!1%Ï!1è%Ï!1[Ò!1´[Ò!1!è!1@!è!1¿("1¼¿("1m/"1Äm/"1÷8"1@÷8"1-<"1 -<"1óQ"1˜óQ"1‘’"1‘’"1?™"1 ?™"1É¢"1˜ É¢"1ÿ¥"1d ÿ¥"1AÓ"1ð AÓ"1‹Ú"1Ì ‹Ú"1ëþ"1H ëþ"1-,#1h-,#1—:#1D—:#1So#1ÐSo#13�#1ô3�#11š#1 1š#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1�€&1à�€&1]R'1$]R'1e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¸ì¤p b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]�f¦ú: bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¸ì¤p b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]�f¦ú: bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¸ì¤p  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]�f¦ú:  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 }�b‚ * bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1]�‡Çú:—×4t  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d]Ý:º—ôt–ÑQÛ[®.  ‹ eåhèEÅ"¢ÿ Ü \ ¹ 9 –  s ó ×  P Ð -mÊ g§D bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSŠŠ)?*?4³4³>'?'I™I™S T ^‰^‰hûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$*%*'ž'ž)*,†,†.ú.ú0¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnñnñpcqcsÕsÕuGvGx¹x¹z+{+}�}�€‚�‚�„ó„ó†e‡e‰׉׋IŒI޻޻�-‘-“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx7i!1¹}!1è¹}!1W¾!1dW¾!1¡Å!1l¡Å!1+Ï!1è+Ï!1aÒ!1´aÒ!1…ç!1@…ç!1#("1¼#("1s/"1Äs/"1ý8"1@ý8"13<"1 3<"1WQ"1˜WQ"1õ‘"1õ‘"1E™"1 E™"1Ï¢"1˜ Ï¢"1¦"1d ¦"1GÓ"1ð GÓ"1Ú"1Ì Ú"1yþ"1H yþ"1»+#1h»+#1�:#1D�:#1Yo#1ÐYo#19�#1ô9�#17š#1 7š#1Õ$1¬Õ$1­ò$1p­ò$1çÞ%1tçÞ%1aì%1Taì%1A�&1àA�&1S'1$S'1¹e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ò b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^žû; bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ò b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^žû; bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ò  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^žû;  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ííf†ñ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1E]�ú:—×4t  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d…]ÝÎN:º——`àôtÑQ®.‹ hèEÅ"¢ÿ Ü \ ¹ 9 –  s ó • Õ P Ð -mÊ g§D bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ø ØPPÈÈ)@*@4¸4¸>0?0I¦I¦SÜSÜ]œ^œhisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ØØPPÈÈ @ @ ¸ ¸00¨¨  ˜˜ˆˆ x x"ð"ð$h%h'à'à)X*X,Ð,Ð.H/H1À1À38486°6°8(9(; ; =>@�@�BCE€E€GøGøIpJpLæLæN\O\QÒQÒSHTHV¾V¾X4Y4[ª[ª] ^ `–`–bÌbÌdBeBg¸g¸i.j.l¤l¤ndodqÚqÚsPtPvÆvÆx?I’I’SÇSÇ]z^zhíhír`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?K@KB¾B¾D1E1G¤G¤IJLŠLŠNýNýPpQpSãSãUVVVXÉXÉZ<[<]¯]¯_"`"bWbWdÊdÊf=g=i°i°k#l#nÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†T‡T‰ƉƋ8Œ8ŽªŽª�‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx7i!1³}!1è³}!1Q¾!1dQ¾!1¡Å!1l¡Å!1+Ï!1è+Ï!1aÒ!1´aÒ!1…ç!1@…ç!1#("1¼#("1s/"1Äs/"1ý8"1@ý8"13<"1 3<"1WQ"1˜WQ"1õ‘"1õ‘"1E™"1 E™"1Ï¢"1˜ Ï¢"1¦"1d ¦"1GÓ"1ð GÓ"1OÚ"1Ì OÚ"1¯þ"1H ¯þ"1ñ+#1hñ+#1�:#1D�:#1Yo#1ÐYo#19�#1ô9�#17š#1 7š#1½$1¬½$1•ò$1p•ò$1ÏÞ%1tÏÞ%1Iì%1TIì%1É€&1àÉ€&1™R'1$™R'1Ae'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ð b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^žû; bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ð b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^žû; bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘Ð  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^žû;  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Íí{›&F bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1]�ŽÎú:—×4t  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d]Ý:º—ôt)©ÑQòr®.»;‹ „hèEÅ"¢ÿ Ü \ ¹ 9 –  s ó P Ð  ^ -mÊ g§D bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ö ÖLLÂÂ)8*84®4®>$?$I˜I˜SËSË]ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††üürrèè^ ^"Ô"Ô$J%J'À'À)6*6,¬,¬."/"1˜1˜346„6„8ú8ú:p;p=æ=æ?\@\BÒBÒDHEHG¾G¾I4J4L¨L¨NOQ�Q�STVxVxXìXìZ`[`]Ô]Ô_H`Hb{b{dïdïfcgci×i×kKlKnoq|q|sðsðudvdxØxØzL{L}À}À4€4‚¨‚¨„…‡�‡�‰ŠŒxŒxŽìŽì�`‘`“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx7i!1I~!1èI~!1ç¾!1dç¾!1¡Å!1l¡Å!1+Ï!1è+Ï!1aÒ!1´aÒ!1è!1@è!1¹("1¼¹("1s/"1Äs/"1ý8"1@ý8"13<"1 3<"1íQ"1˜íQ"1‹’"1‹’"1E™"1 E™"1Ï¢"1˜ Ï¢"1¦"1d ¦"1GÓ"1ð GÓ"1+Ú"1Ì +Ú"1‹þ"1H ‹þ"1Í+#1hÍ+#1Á:#1DÁ:#1}o#1Ð}o#1í�#1ôí�#1ëš#1 ëš#1Ÿ$1¬Ÿ$1wò$1pwò$1±Þ%1t±Þ%1+ì%1T+ì%1!€&1à!€&1ñQ'1$ñQ'1™d'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1­á™e b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]�®îú: bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1­á™e b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]�®îú: bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1­á™e  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]�®îú:  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 x˜Íí  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1g§ŽÎN³óX˜  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dgçNÎ5µœ8¸ƒ†êjÔTÑQ¢"¸8�‚ g ç L Ì 1 ±  – û { ^ ž à `Çn®S¸ø bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØKK¾¾)1*14¤4¤>?I†I†SºSº]l^lhÝhÝrNsN} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆûûnnááT T"Ç"Ç$:%:'­'­) * ,“,“./1y1y3ì3ì5_6_8Ò8Ò:E;E=¸=¸?)@)BšBšD E G|G|IíIíK^L^NÏNÏP@Q@S±S±U"V"X“X“Z[]u]u_æ_æabd‹d‹füfühmimkÞkÞm�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1¹}!1й}!1W¾!1LW¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1‹ç!1(‹ç!1)("1¤)("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1]Q"1€]Q"1û‘"1üû‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1%Ú"1´ %Ú"1…þ"10 …þ"1Ç+#1PÇ+#1i2#1,i2#1£‹#1˜£‹#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ÿ€&1Ìÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ó b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U]�ú: bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ó b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U]�ú: bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ó  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U]�ú:  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ííg‡õ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1B€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1‚ÉI�W×r ²  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` × ×NNÅÅ)<*<4³4³>*?*IŸIŸST^‰^‰hþhþrsss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``××NNÅÅ < < ³ ³**¡¡��}}ôôk k"â"â$Y%Y'Ð'Ð)G*G,¾,¾.5/51¬1¬3#4#6š6š89;ˆ;ˆ=ÿ=ÿ?v@vBíBíDdEdGÛGÛIRJRLÇLÇN bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 ÔŠV b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]�y¹þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 ÔŠV  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]�y¹þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ŸÍí: bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÙY²2‹ € À  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝRRÇÇ)<*<4±4±>'?'I›I›ST^ƒ^ƒh÷h÷rmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��‡‡üüq q"æ"æ$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡�‡�‰ŠŒwŒwŽìŽì�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1¹}!1й}!1W¾!1LW¾!1¡Å!1T¡Å!1+Ï!1Ð+Ï!1aÒ!1œaÒ!1‹ç!1(‹ç!1)("1¤)("1s/"1¬s/"1ý8"1(ý8"13<"1ô3<"1]Q"1€]Q"1û‘"1üû‘"1E™"1 E™"1Ï¢"1€ Ï¢"1¦"1L ¦"1GÓ"1Ø GÓ"1[Ú"1´ [Ú"1»þ"10 »þ"1ý+#1Pý+#1ñ4#1,ñ4#1-Œ#1˜-Œ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ñ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V]�ú: bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ñ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V]�ú: bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ñ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V]�ú:  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Íí~ž,L bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1‰É€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1žåe¬,sóù 9  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×MMÂÂ)7*74¬4¬>!?!I”I”ST^z^zhíhír`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b£b£deg‰g‰iüiükolonânâpUqUsÈsÈu;v;x®x®z!{!}”}”€‚z‚z„í„í†`‡`‰Ó‰Ó‹FŒF޹޹�,‘,“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l7i!1C~!1ÐC~!1á¾!1Lá¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1è!1(è!1³("1¤³("1y/"1¬y/"19"1(9"19<"1ô9<"1çQ"1€çQ"1…’"1ü…’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1Ú"1´ Ú"1sþ"10 sþ"1µ+#1Pµ+#1ý4#1,ý4#1M�#1˜M�#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1‡€&1̇€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖŽZ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]�y¹ÿ? bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖŽZ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]�y¹ÿ? bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖŽZ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]�y¹ÿ?  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 k‹Ííæ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1™Ù€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1dä=½–ïoY €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84­4­>"?"I•I•ST^y^yhëhër_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?N@NBÃBÃD8E8G­G­I"J"L•L•NOQ{Q{SîSîUaVaXÔXÔZG[G]º]º_-`-bŸbŸdegƒgƒiõiõkglgnÙnÙpKqKs½s½u/v/x¡x¡z{}…}…÷÷�i‚i„Û„Û†O‡O‰‰‹5Œ5ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¿}!1п}!1]¾!1L]¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1‘ç!1(‘ç!1/("1¤/("1y/"1¬y/"19"1(9"19<"1ô9<"1cQ"1€cQ"1’"1ü’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#1i2#1,i2#1EŒ#1˜EŒ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1á€&1Ìá€&1±R'1±R'1Ye'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ô b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @_Ÿü< bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ô b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @_Ÿü< bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ô  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @_Ÿü<  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ^~åk‹ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÐP  pð P  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Þ ÞTTÊÊ)@*@4¶4¶>*?*IžIžST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÞÞXXÒÒ L L Æ Æ@@ºº44®®$$šš † †"ü"ü$r%r'è'è)^*^,Ô,Ô.J/J1À1À36466¬6¬8"9";˜;˜=>@‚@‚BöBöDjEjGÞGÞIRJRLÆLÆN:O:Q®Q®S"T"V–V–X Y [~[~]ò]ò_f`fbÚbÚdNeNgÂgÂi6j6lªlªnoq’q’stvzvzxîxîzb{b}Ö}ÖJ€J‚¾‚¾„2…2‡¦‡¦‰ŠŒŒŒŒŽÿŽÿ�r‘r“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!17~!1Ð7~!1Õ¾!1LÕ¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1 è!1( è!1§("1¤§("1…/"1¬…/"19"1(9"1E<"1ôE<"1ÛQ"1€ÛQ"1y’"1üy’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"17Ú"1´ 7Ú"1—þ"10 —þ"1Ù+#1PÙ+#1i2#1,i2#1e�#1˜e�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1ù€&1Ìù€&1ÉR'1ÉR'1qe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•ÉK b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]�¤ä I bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•ÉK b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]�¤ä I bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•ÉK  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]�¤ä I  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Íív–$ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ädÈH¬,0p €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>1?1I¨I¨ST^‘^‘hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIYJYLÐLÐNGOGQ¾Q¾S5T5V¬V¬X#Y#[š[š]^`ˆ`ˆbÿbÿdvevgígíidjdlÛlÛnMoMq¿q¿s1t1v£v£xy{‡{‡}ù}ùk€k‚݂݄O…O‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¿}!1п}!1]¾!1L]¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1‘ç!1(‘ç!1/("1¤/("1y/"1¬y/"19"1(9"19<"1ô9<"1cQ"1€cQ"1’"1ü’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1Ú"1´ Ú"1ßþ"10 ßþ"1!,#1P!,#1÷4#1,÷4#1ÕŒ#1˜ÕŒ#1±$1$±$1‰ò$1è‰ò$1ÃÞ%1ìÃÞ%1i€&1Ìi€&19R'19R'1ád'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ò b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^žû; bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ò b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^žû; bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“Ò  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^žû;  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 w—Jj8 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Ѐ� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@À�à`°0  à  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜPPÄÄ)8*84¬4¬>?I�I�ST^t^thæhærXsX} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœ„„øøl l"à"à$T%T'È'È)<*<,°,°.$/$1˜1˜3 4 6€6€8ô8ô:h;h=Ü=Ü?N@NBÀBÀD2E2G¤G¤IJLˆLˆNúNúPlQlSÞSÞUPVPXÂXÂZ4[4]¦]¦_`bŠbŠdüdüfngniàiàkRlRnÄnÄp6q6s¨s¨uvxŒxŒzþzþ|p}pââ�T‚T„ƄƆ8‡8‰©‰©‹ŒŽ‹Ž‹�ü�ü’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1=~!1Ð=~!1Û¾!1LÛ¾!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1 è!1( è!1§("1¤§("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1ÛQ"1€ÛQ"1y’"1üy’"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1ûÙ"1´ ûÙ"1[þ"10 [þ"1�+#1P�+#15#1,5#1Ž#1˜Ž#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1Ÿ€&1ÌŸ€&1oR'1oR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—˃O b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]�£ã J bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—˃O b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]�£ã J bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—˃O  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]�£ã J  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ^~ÍíÌì bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¤ä€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�tôXØ<¼Ô €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ø ØMMÂÂ)7*74¬4¬>#?#I˜I˜S T ^^hñhñrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccØØMM 7 7 ¬ ¬!!––  €€õõjjßßT T"É"É$>%>'³'³)(*(,�,�./1‡1‡3ü3ü5q6q8æ8æ:[;[=Ð=Ð?E@EBºBºD/E/G¤G¤IJL�L�NOQzQzSïSïUdVdXÙXÙZN[N]Ã]Ã_8`8b­b­d"e"g—g—i j l�l�nónópeqes×s×uIvIx»x»z-{-}Ÿ}Ÿ€‚ƒ‚ƒ„õ„õ†g‡g‰ىًKŒK޽޽�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1Å}!1ÐÅ}!1c¾!1Lc¾!1ÅÅ!1TÅÅ!1OÏ!1ÐOÏ!1…Ò!1œ…Ò!1—ç!1(—ç!15("1¤5("1—/"1¬—/"1!9"1(!9"1W<"1ôW<"1iQ"1€iQ"1’"1ü’"1i™"1 i™"1ó¢"1€ ó¢"1)¦"1L )¦"1kÓ"1Ø kÓ"1ÁÚ"1´ ÁÚ"1!ÿ"10 !ÿ"1c,#1Pc,#1o2#1,o2#1ùŒ#1˜ùŒ#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%15�&1Ì5�&1S'1S'1­e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” Õ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  IuµR bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” Õ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  IuµR bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” Õ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  IuµR  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 PpÐð†¦ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€Û[¶6‘À  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ý ÝVVÏÏ)H*H4Á4Á>:?:I±I±S&T&^™^™h i ss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÝÝVVÏÏ H H Á Á::³³,,¥¥—— ‰ ‰"#%{%{'ô'ô)m*m,æ,æ._/_1Ø1Ø3Q4Q6Ê6Ê8C9C;¼;¼=5>5@®@®B'C'E E GHJ’J’L M O€O€Q÷Q÷SnTnVåVåX\Y\[Ó[Ó]J^J`Á`Áb6c6e«e«g h j•j•l m o}o}qðqðsctcvÖvÖxIyI{¼{¼}/~/€¢€¢‚ƒ…ˆ…ˆ‡û‡û‰mŠmŒߌߎQ�Q‘ÑÓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!11~!1Ð1~!1Ͼ!1LϾ!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1è!1(è!1¡("1¤¡("1/"1¬/"1 9"1( 9"1?<"1ô?<"1ÕQ"1€ÕQ"1s’"1üs’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1%Ú"1´ %Ú"1…þ"10 …þ"1Ç+#1PÇ+#1o2#1,o2#1Ž#1˜Ž#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оt@ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]�y¹ý= bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оt@ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]�y¹ý= bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оt@  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]�y¹ý=  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Íía�î bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ïoÞ^ÍMà  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii ã ã]]ÓÓ)I*I4¿4¿>5?5I©I©ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiãã]]×× Q Q Ë ËEE¿¿99³³--§§!! › ›"#%‹%‹'(*w*w,í,í.c/c1Ù1Ù3O4O6Å6Å8;9;;±;±='>'@�@�BCE‰E‰GÿGÿIuJuLéLéN]O]QÑQÑSETEV¹V¹X-Y-[¡[¡]^`‰`‰býbýdqeqgågåiYjYlÍlÍnAoAqµqµs)t)v�v�xy{…{…}ù}ùm€m‚á‚á„U…U‡ɇɉ=Š=Œ±Œ±Ž%�%‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1Å}!1ÐÅ}!1c¾!1Lc¾!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1‘ç!1(‘ç!1/("1¤/("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1cQ"1€cQ"1’"1ü’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1eÓ"1Ø eÓ"1©Ú"1´ ©Ú"1 ÿ"10  ÿ"1K,#1PK,#1 5#1, 5#1•�#1˜•�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1À&1ÌÀ&1“R'1“R'1;e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ó b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jt´Q bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ó b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jt´Q bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`”Ó  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jt´Q  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 p�&o� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1›Û€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1lìGÇ"¢ý}[› €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I£I£ST^‡^‡høhørjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbæbædYeYgÌgÌi?j?l²l²n#o#q”q”stvvvvxçxçzX{X}É}É:€:‚«‚«„…‡އމÿ‰ÿ‹pŒpŽáŽá�R‘R“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!17~!1Ð7~!1Õ¾!1LÕ¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1è!1(è!1¡("1¤¡("1/"1¬/"1 9"1( 9"1?<"1ô?<"1ÕQ"1€ÕQ"1s’"1üs’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1õÙ"1´ õÙ"1Uþ"10 Uþ"1—+#1P—+#15#1,5#1ÇŽ#1˜ÇŽ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀxD b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]�y¹þ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀxD b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]�y¹þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀxD  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]�y¹þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÍíQq½Ý bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¯ï€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¼<«+š‰ �Ï €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßWWËË)?*?4³4³>'?'I™I™S T ^}^}hïhïrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ��  "÷"÷$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1Ë}!1ÐË}!1i¾!1Li¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1�ç!1(�ç!1;("1¤;("1/"1¬/"1 9"1( 9"1?<"1ô?<"1oQ"1€oQ"1 ’"1ü ’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1ÁÚ"1´ ÁÚ"1!ÿ"10 !ÿ"1c,#1Pc,#1o2#1,o2#1­�#1˜­�#1á$1$á$1¹ò$1è¹ò$1óÞ%1ìóÞ%1G�&1ÌG�&1S'1S'1¿e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a• Ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^žû; bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a• Ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^žû; bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a• Ö  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^žû;  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 IiÂâ�¡ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€æfÌL²2p° €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜTTÌÌ)D*D4¼4¼>4?4IªIªS T ^–^–h i s‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœŒŒ | |"ô"ô$l%l'ä'ä)\*\,Ô,Ô.L/L1Ä1Ä3<4<6´6´8,9,;¤;¤=>@”@”B C E„E„GüGüItJtLêLêN`O`QÖQÖSLTLVÂVÂX8Y8[®[®]$^$`š`šbce†e†gügüirjrlèlèn^o^qÔqÔsJtJvÀvÀx6y6{¬{¬}"~"€˜€˜‚ƒ…„…„‡ú‡ú‰pŠpŒæŒæŽ\�\‘Ò‘Ò“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1+~!1Ð+~!1ɾ!1Lɾ!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1÷ç!1(÷ç!1•("1¤•("1/"1¬/"1 9"1( 9"1?<"1ô?<"1ÉQ"1€ÉQ"1g’"1üg’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1»Ú"1´ »Ú"1ÿ"10 ÿ"1],#1P],#1o2#1,o2#1ÍŽ#1˜ÍŽ#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³i5 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]��Ýû; bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³i5 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]��Ýû; bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1³i5  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]��Ýû;  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 d„Øøzš bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€úzôtîn�Ð €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜQQÆÆ);*;4°4°>#?#I–I–S T ^|^|hïhïrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒvvëë` `"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?O@OBÂBÂD5E5G¨G¨IJLŽLŽNOQtQtSçSçUZVZXÍXÍZ@[@]³]³_&`&b™b™d e ggiòiòkelenØnØpKqKs¾s¾u1v1x¤x¤z{}Š}Šýý�p‚p„ã„ã†V‡V‰ȉȋ:Œ:ެެ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1Ë}!1ÐË}!1i¾!1Li¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1—ç!1(—ç!15("1¤5("1/"1¬/"1 9"1( 9"1?<"1ô?<"1iQ"1€iQ"1’"1ü’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1µÚ"1´ µÚ"1ÿ"10 ÿ"1W,#1PW,#15#1,5#1UŽ#1˜UŽ#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1Õ€&1ÌÕ€&1¥R'1¥R'1Me'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ô b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^žû; bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ô b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^žû; bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ô  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^žû;  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 i‰ôv– bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¦æ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1˜~þdäJÊV €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÆÆ)<*<4²4²>(?(IœIœST^„^„høhørlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$N%N'Ä'Ä):*:,°,°.&/&1œ1œ346ˆ6ˆ8þ8þ:t;t=ê=ê?`@`BÖBÖDLELGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡Œ‡Œ‰ŠŒtŒtŽèŽè�\‘\“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1+~!1Ð+~!1ɾ!1Lɾ!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1ýç!1(ýç!1›("1¤›("1/"1¬/"1 9"1( 9"1?<"1ô?<"1ÏQ"1€ÏQ"1m’"1üm’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1£Ú"1´ £Ú"1ÿ"10 ÿ"1E,#1PE,#1-5#1,-5#1‡�#1˜‡�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%13€&1Ì3€&1R'1R'1«d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µm9 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]��Ýû; bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µm9 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]��Ýû; bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µm9  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]��Ýû;  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Dd§Çe… bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ºú€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1èhâbÜ\ÖVJŠ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜOOÂÂ)5*54¨4¨>?IŠIŠSûSû]l^lhÝhÝrNsN} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒÿÿrrååX X"Ë"Ë$>%>'±'±)$*$,—,—. / 1}1}3ð3ð5c6c8Ö8Ö:I;I=¼=¼?-@-BžBžDEG€G€IñIñKbLbNÓNÓPDQDSµSµU&V&X—X—Z[]y]y_ê_êa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1Ñ}!1ÐÑ}!1o¾!1Lo¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1�ç!1(�ç!1;("1¤;("1…/"1¬…/"19"1(9"1E<"1ôE<"1oQ"1€oQ"1 ’"1ü ’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1»Ú"1´ »Ú"1ÿ"10 ÿ"1],#1P],#1u2#1,u2#1aŽ#1˜aŽ#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1�&1Ì�&1áR'1áR'1‰e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– × b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^žû; bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– × b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^žû; bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– ×  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^žû;  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Bb´Ôzš bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ñqâbÓS ` €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I£I£ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbèbèd]e]gÒgÒiGjGl¼l¼n1o1q¦q¦stv�v�xy{z{z}ï}ïd€d‚قلN…N‡ÇÉ8Š8Œ­Œ­Ž"�"‘—‘—“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1%~!1Ð%~!1þ!1Lþ!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1ñç!1(ñç!1�("1¤�("1…/"1¬…/"19"1(9"1E<"1ôE<"1ÃQ"1€ÃQ"1a’"1üa’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1u2#1,u2#1��#1˜��#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1á€&1Ìá€&1±R'1±R'1Ye'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨^* b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]�{»ú: bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨^* b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]�{»ú: bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨^*  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]�{»ú:  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Ww¾Þ‰© bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€… Š�@€ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áVVËË)@*@4µ4µ>+?+IŸIŸST^‡^‡hûhûrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‹‹ u u"ê"ê$_%_'Ô'Ô)I*I,¾,¾.3/31¨1¨346’6’89;|;|=ñ=ñ?f@fBÛBÛDPEPGÅGÅI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÃbÃd7e7g«g«ijl“l“noq{q{sïsïucvcx×x×zK{K}¿}¿3€3‚§‚§„…‡‘‡‘‰ŠŒ{Œ{ŽðŽð�e‘e“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1Ë}!1ÐË}!1i¾!1Li¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1�ç!1(�ç!1;("1¤;("1…/"1¬…/"19"1(9"1E<"1ôE<"1oQ"1€oQ"1 ’"1ü ’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1»Ú"1´ »Ú"1ÿ"10 ÿ"1],#1P],#1'5#1,'5#1�#1˜�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1«€&1Ì«€&1{R'1{R'1#e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– Õ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^žû; bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– Õ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^žû; bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– Õ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^žû;  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 b‚âw— bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1±ñ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÄDµ5¦&—Ñ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÆÆ);*;4°4°>%?%I˜I˜S T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnænæpYqYsÌsÌu?v?x²x²z%{%}˜}˜ € ‚~‚~„ñ„ñ†d‡d‰׉׋JŒJ޽޽�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1%~!1Ð%~!1þ!1Lþ!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1÷ç!1(÷ç!1•("1¤•("1…/"1¬…/"19"1(9"1E<"1ôE<"1ÉQ"1€ÉQ"1g’"1üg’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1ÇÚ"1´ ÇÚ"1'ÿ"10 'ÿ"1i,#1Pi,#195#1,95#1G�#1˜G�#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªb. b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]�zºú: bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªb. b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]�zºú: bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªb.  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]�zºú:  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 7W‘±†¦ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Å€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1”™ž#£E €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝRRÇÇ)<*<4±4±>&?&I™I™S T ^}^}hïhïrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��wwììa a"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD.?.I¢I¢ST^Š^Šhþhþrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²((žž Š Š"#%v%v'ì'ì)b*b,Ø,Ø.N/N1Ä1Ä3:4:6°6°8&9&;œ;œ=>@†@†BúBúDnEnGâGâIVJVLÊLÊN>O>Q²Q²S&T&VšVšXY[‚[‚]ö]ö_j`jbÞbÞdReRgÆgÆi:j:l®l®n"o"q–q–s t v~v~xòxòzf{f}Ú}ÚN€N‚‚„6…6‡ª‡ª‰ŠŒ�Œ�Ž�‘v‘v“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lOi!1~!1Ð~!1·¾!1L·¾!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1ëç!1(ëç!1‰("1¤‰("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1½Q"1€½Q"1[’"1ü[’"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1{2#1,{2#15�#1˜5�#1á$1$á$1¹ò$1è¹ò$1óÞ%1ìóÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�S b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]��Ýû; bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�S b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]��Ýû; bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�S  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]��Ýû;  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Jj¤Ä„¤ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€�  0°ð0 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii à àWWÎÎ)E*E4¼4¼>5?5I¬I¬S#T#^•^•hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiààWWÎÎ E E ¼ ¼33ªª!!˜˜††ýýt t"ë"ë$b%b'Ù'Ù)P*P,Ç,Ç.>/>1µ1µ3,4,6£6£89;‘;‘=>@@BöBöDmEmGäGäI]J]LÔLÔNKOKQÂQÂS9T9V°V°X'Y'[ž[ž]^`Œ`ŒbcezezgñgñihjhlßlßnQoQqÃqÃs5t5v§v§xy{‹{‹}ý}ýo€o‚á‚á„S…S‡Ňʼn7Š7Œ©Œ©Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lOi!1Ñ}!1ÐÑ}!1o¾!1Lo¾!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1£ç!1(£ç!1A("1¤A("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1uQ"1€uQ"1’"1ü’"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1Ú"1´ Ú"1sþ"10 sþ"1µ+#1Pµ+#135#1,35#1Ï�#1˜Ï�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1�€&1Ì�€&1QR'1QR'1ùd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c— Ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @^žû; bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c— Ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @^žû; bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c— Ö  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @^žû;  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Íí[{Ðð bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¼ü€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ðpìlèhädŒÌ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àTTÈÈ)<*<4°4°>"?"I”I”ST^x^xhêhêr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  ˆˆüüp p"ä"ä$X%X'Ì'Ì)@*@,´,´.(/(1œ1œ346„6„8ø8ø:l;l=à=à?R@RBÄBÄD6E6G¨G¨IJLŒLŒNþNþPpQpSâSâUTVTXÆXÆZ8[8]ª]ª_`bŽbŽdegrgriäiäkVlVnÈnÈp:q:s¬s¬uvx�x�z{}t}tææ�X‚X„ʄʆ<‡<‰­‰­‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lOi!1~!1Ð~!1½¾!1L½¾!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1ñç!1(ñç!1�("1¤�("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1ÃQ"1€ÃQ"1a’"1üa’"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1K5#1,K5#1‘#1˜‘#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1±€&1̱€&1�R'1�R'1)e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸW# b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 ]��Ýû; bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸW# b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 ]��Ýû; bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸW#  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 ]��Ýû;  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 *J{›…¥ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ѐ� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@ÀPÐ`àpðÀ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ü ÜQQÆÆ);*;4°4°>'?'IœIœST^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÜÜQQÆÆ ; ; ° °%%šš„„ùùnnããX X"Í"Í$B%B'·'·),*,,¡,¡./1‹1‹346u6u8ê8ê:_;_=Ô=Ô?I@IB¾B¾D3E3G¨G¨IJL”L”N O Q~Q~SóSóUhVhXÝXÝZR[R]Ç]Ç_<`M?MIÄIÄS9T9^¬^¬his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðyyòòkkää ] ] Ö ÖOOÈÈAAºº33¬¬%% ž ž"#%�%�' ( *‚*‚,û,û.t/t1í1í3f4f6ß6ß8X9X;Ñ;Ñ=J>J@Ã@ÃB@?@I´I´S(T(^œ^œhis„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvððjjää ^ ^ Ø ØRRÌÌFFÀÀ::´´.. ¨ ¨""#"%˜%˜'(*„*„,ú,ú.p/p1æ1æ3\4\6Ò6Ò8H9H;¾;¾=4>4@ª@ªB C E–E–G H J€J€LôLôNhOhQÜQÜSPTPVÄVÄX8Y8[¬[¬] ^ `”`”bce|e|gðgðidjdlØlØnLoLqÀqÀs4t4v¨v¨xy{�{�}~€x€x‚ì‚ì„`…`‡Ô‡Ô‰HŠHŒ¼Œ¼Ž0�0‘¤‘¤“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1é}!1�é}!1‡¾!1‡¾!1¿Â!1$¿Â!1»ç!1�»ç!1Y("1Y("1‘,"1$‘,"1�Q"1��Q"1+’"1+’"1c–"1$c–"1Ù"1�Ù"1}ý"1ü}ý"1;0#1 ;0#1#1| #1ç$1Ü ç$1¿ò$1  ¿ò$1ùÞ%1¤ ùÞ%1ç€&1„ ç€&1·R'1È·R'1_e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14s§æÒž b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14s§æÒž b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14s§æÒž b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Íí  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ww î îeeÜÜ)S*S4Ê4Ê>???I´I´S'T'^˜^˜h i sysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðwwîîeeÜÜ S S Ê ÊAA¸¸//¦¦””  ‚ ‚"ù"ù$p%p'ç'ç)^*^,Õ,Õ.L/L1Ã1Ã3:4:6±6±8(9(;Ÿ;Ÿ=>@�@�BCE{E{GòGòIgJgLÜLÜNQOQQÆQÆS;T;V°V°X%Y%[š[š]^`„`„b÷b÷djejgÝgÝiPjPlÃlÃn4o4q¥q¥stv‡v‡xøxøzi{i}Ú}ÚK€K‚¼‚¼„-…-‡�‡�‰ŠŒŒŽðŽð�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1~!1�~!1·¾!1 ·¾!1MÅ!1MÅ!1ëç!1€ëç!1‰("1ü‰("1/"1/"1½Q"1p½Q"1[’"1ì[’"1ñ˜"1ôñ˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1;0#1ì;0#1#1L #1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1·€&1T ·€&1‡R'1˜‡R'1/e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘I b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ѐ} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘I b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ѐ} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1]‘I b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ð €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1y™  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv î îffÚÚ)N*N4Â4Â>4?4I¦I¦ST^Š^Šhühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvîîffÞÞ V V Î ÎFF¾¾66®®&&žž Ž Ž"#%z%z'î'î)b*b,Ö,Ö.J/J1¾1¾32426¦6¦89;Ž;Ž=>@v@vBêBêD^E^GÒGÒIDJDL¶L¶N(O(QšQšS T V~V~XðXðZb[b]Ô]Ô_F`Fb¸b¸d*e*gœgœijl€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1é}!1�é}!1‡¾!1 ‡¾!1¹Â!1¹Â!1µç!1€µç!1S("1üS("1‹,"1‹,"1‡Q"1p‡Q"1%’"1ì%’"1]–"1ô]–"1Ù"1`Ù"1wý"1Ìwý"150#1ì50#1#1L #1ÿ$1¬ ÿ$1×ò$1p ×ò$1ß%1t ß%1e�&1T e�&15S'1˜5S'1Ýe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦ç b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦ç b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦ç b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Íí  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10uu í íeeÝÝ)U*U4Í4Í>C?CI¹I¹S/T/^¥^¥his‘s‘} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðuuííeeÝÝ U U Í ÍEE½½55­­%%�� � �"#%}%}'õ'õ)m*m,å,å.]/]1Õ1Õ3M4M6Å6Å8=9=;µ;µ=->-@¥@¥BCE•E•G H JƒJƒLùLùNoOoQåQåS[T[VÑVÑXGYG[½[½]3^3`©`©bce•e•g h j�j�l÷l÷nmomqãqãsYtYvÏvÏxEyE{»{»}1~1€§€§‚ƒ…“…“‡ ˆ ŠŠŒõŒõŽk�k‘á‘á“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1½¾!1 ½¾!1YÅ!1YÅ!1ëç!1€ëç!1‰("1ü‰("1+/"1+/"1½Q"1p½Q"1[’"1ì[’"1ý˜"1ôý˜"1Ù"1`Ù"1wý"1Ìwý"150#1ì50#1#1L #1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšP b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ý€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšP b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ý€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšP b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ý €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Íí  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo å åZZÏÏ)D*D4·4·>*?*I�I�ST^ƒ^ƒhöhörgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooåå[[ÑÑ G G ½ ½33©©••  ôôi i"Þ"Þ$S%S'È'È)=*=,²,².'/'1œ1œ346†6†8û8û:p;p=ã=ã?V@VBÉBÉD5?5I©I©ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðsséé__ÕÕ K K Á Á77­­##™™……ûûq q"ç"ç$]%]'Ó'Ó)I*I,¿,¿.5/51«1«3!4!6—6—8 9 ;ƒ;ƒ=ù=ù?o@oBåBåD[E[GÑGÑIEJEL¹L¹N-O-Q¡Q¡STV‰V‰XýXýZq[q]å]å_Y`YbÍbÍdAeAgµgµi)j)l�l�noq…q…sùsùumvmxáxázU{U}É}É=€=‚±‚±„%…%‡™‡™‰ Š Œ�Œ�ŽõŽõ�i‘i“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1~!1�~!1½¾!1 ½¾!1YÅ!1YÅ!1ñç!1€ñç!1�("1ü�("1+/"1+/"1ÃQ"1pÃQ"1a’"1ìa’"1ý˜"1ôý˜"1Ù"1`Ù"1wý"1Ìwý"150#1ì50#1#1L #1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1E€&1T E€&1R'1˜R'1½d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšR b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ý€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšR b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ý€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšR b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ý €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Íí  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ç çZZÍÍ)@*@4±4±>"?"I“I“ST^u^uhæhærUsU} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqçç]]ÓÓ I I ¿ ¿55««!!——  }}ððc c"Ö"Ö$I%I'¼'¼)/*/,¢,¢./1ˆ1ˆ3û3û5n6n8á8á:T;T=Å=Å?6@6B§B§DEG‰G‰IúIúKkLkNÜNÜPMQMS¾S¾U/V/X X Z[]‚]‚_ó_óadbddÕdÕfFgFi·i·k(l(n™n™p q s{s{uìuìw]x]zÎzÎ|?}?°°�!‚!„’„’†‡‰q‰q‹á‹á�QŽQ�Á�Á’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1ã}!1�ã}!1�¾!1 �¾!1¿Â!1¿Â!1¯ç!1€¯ç!1M("1üM("1‘,"1‘,"1�Q"1p�Q"1’"1ì’"1c–"1ôc–"1Ù"1`Ù"1qý"1Ìqý"150#1ì50#1#1L #1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1#�&1T #�&1óR'1˜óR'1›e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£ä b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£ä b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£ä b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Íí  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq è è__ÖÖ)M*M4Ä4Ä>9?9I®I®S#T#^˜^˜h i s‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqèè__ÖÖ M M Ä Ä;;²²))  ŽŽ | |"ó"ó$j%j'á'á)X*X,Ï,Ï.F/F1½1½34446«6«8"9";™;™=>@‡@‡BþBþDuEuGìGìIaJaLÖLÖNKOKQÀQÀS5T5VªVªXY[”[”] ^ `~`~bóbódhehgÝgÝiRjRlÇlÇn.?.I¢I¢ST^Š^Šhþhþrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnææ^^ÖÖ N N Æ Æ>>¶¶..¦¦�� z z"ï"ï$d%d'Ù'Ù)N*N,Ã,Ã.8/81­1­3"4"6—6—8 9 ;�;�=ö=ö?k@kBàBàDUEUGÊGÊI>J>L²L²N&O&QšQšSTV‚V‚XöXöZj[j]Þ]Þ_R`RbÆbÆd:e:g®g®i"j"l–l–n o q~q~sòsòufvfxÚxÚzN{N}Â}Â6€6‚ª‚ª„…‡’‡’‰ŠŒ|Œ|ŽñŽñ�f‘f“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1Ý}!1�Ý}!1{¾!1 {¾!1ÅÂ!1ÅÂ!1¯ç!1€¯ç!1M("1üM("1—,"1—,"1�Q"1p�Q"1’"1ì’"1i–"1ôi–"1Ù"1`Ù"1qý"1Ìqý"150#1ì50#1#1L #1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1·€&1T ·€&1‡R'1˜‡R'1/e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡à b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡à b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡à b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Íí  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æ\\ÑÑ)F*F4»4»>.?.I¡I¡ST^‡^‡húhúrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  ‚‚øøn n"ä"ä$Y%Y'Î'Î)C*C,¸,¸.-/-1¢1¢346Œ6Œ89;v;v=ë=ë?`@`BÕBÕDJEJG¿G¿I2J2L¥L¥NOQ‹Q‹SþSþUqVqXäXäZW[W]Ê]Ê_=`=b°b°d#e#g–g–i j l|l|nïnïpbqbsÕsÕuHvHx»x»z.{.}¡}¡€‚‡‚‡„ú„ú†m‡m‰à‰à‹SŒSŽÆŽÆ�9‘9“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1%~!1�%~!1þ!1 þ!1/Å!1/Å!1ñç!1€ñç!1�("1ü�("1/"1/"1ÃQ"1pÃQ"1a’"1ìa’"1Ó˜"1ôÓ˜"1Ù"1`Ù"1qý"1Ìqý"150#1ì50#1#1L #1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£[' b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1y¹€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£[' b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1y¹€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£[' b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1y¹ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Íí  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn ä äYYÎÎ)C*C4¸4¸>+?+IžIžST^‚^‚hôhôrfsf} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnääZZÐÐ F F ¼ ¼22¨¨””  ~~óóh h"Ý"Ý$R%R'Ç'Ç)<*<,±,±.&/&1›1›346…6…8ú8ú:o;o=ä=ä?Y@YBÎBÎDCECG¸G¸I+J+LžLžNOQ„Q„S÷S÷UjVjXÝXÝZP[P]Ã]Ã_6`6b¨b¨degŒgŒiþiþkplpnânâpTqTsÆsÆu8v8xªxªz{}Ž}Ž€‚r‚r„ä„ä†V‡V‰ɉɋ<Œ<ޝޝ�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1×}!1�×}!1u¾!1 u¾!1³Â!1³Â!1©ç!1€©ç!1G("1üG("1…,"1…,"1{Q"1p{Q"1’"1ì’"1W–"1ôW–"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1/0#1ì/0#1û~#1L û~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1ó€&1T ó€&1ÃR'1˜ÃR'1ke'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l á b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l á b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l á b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1f†  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ë ëaa××)M*M4Á4Á>5?5I©I©ST^‘^‘hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqëëeeßß Y Y Ó ÓMMÇÇAA»»11§§ “ “" # %%'õ'õ)k*k,á,á.W/W1Í1Í3C4C6¹6¹8/9/;¥;¥=>@�@�BCEuEuGéGéI]J]LÑLÑNEOEQ¹Q¹S-T-V¡V¡XY[‰[‰]ý]ý_q`qbåbådYeYgÍgÍiAjAlµlµn)o)q�q�stv…v…xùxùzm{m}á}áU€U‚ɂɄ=…=‡¯‡¯‰"Š"Œ•Œ•Ž�‘{‘{“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1%~!1�%~!1þ!1 þ!1YÅ!1YÅ!1÷ç!1€÷ç!1•("1ü•("1+/"1+/"1ÉQ"1pÉQ"1g’"1ìg’"1ý˜"1ôý˜"1Ù"1`Ù"1qý"1Ìqý"1/0#1ì/0#1û~#1L û~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬b. b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¦æ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬b. b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¦æ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬b. b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¦æ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Íí  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj á áXXÏÏ)F*F4½4½>4?4I«I«S"T"^”^”hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjááXXÏÏ F F ½ ½44««""™™‡‡þþu u"ì"ì$c%c'Ú'Ú)Q*Q,È,È.?/?1¶1¶3-4-6¤6¤89;’;’= > @€@€B÷B÷DnEnGåGåI\J\LÓLÓNJOJQÁQÁS8T8V¯V¯X&Y&[�[�]^`‹`‹bceyeygðgðigjglÞlÞnPoPqÂqÂs4t4v¦v¦xy{Š{Š}ü}ün€n‚à‚à„R…R‡ćĉ6Š6Œ¨Œ¨Ž�‘Œ‘Œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1×}!1�×}!1u¾!1 u¾!1³Â!1³Â!1£ç!1€£ç!1A("1üA("1…,"1…,"1uQ"1puQ"1’"1ì’"1W–"1ôW–"1µÚ"1`µÚ"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1{€&1T {€&1KR'1˜KR'1ód'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžÝ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžÝ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžÝ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1s“  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo ç ç[[ÏÏ)C*C4µ4µ>'?'I™I™S T ^}^}hïhïr_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooçç__×× O O Ç Ç??··//§§�� w w"ë"ë$_%_'Ó'Ó)G*G,»,».///1£1£346‹6‹8ÿ8ÿ:s;s=å=å?W@WBÉBÉD;E;G­G­IJL‘L‘NOQuQuSçSçUYVYXËXËZ=[=]¯]¯_!`!b“b“degwgwiéiék[l[nÍnÍp?q?s±s±u#v#x•x•z{}y}yëë�]‚]„τφ?‡?‰°‰°‹!Œ!Ž’Ž’�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1+~!1�+~!1ɾ!1 ɾ!1YÅ!1YÅ!1÷ç!1€÷ç!1•("1ü•("1+/"1+/"1ÉQ"1pÉQ"1g’"1ìg’"1ý˜"1ôý˜"1Ù"1`Ù"1qý"1Ìqý"1/0#1ì/0#1û~#1L û~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1±€&1T ±€&1�R'1˜�R'1)e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬d0 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1§ç€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬d0 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1§ç€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬d0 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1§ç €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Íí  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ß ßTTÉÉ)>*>4³4³>(?(I�I�ST^„^„höhörhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjßßTTÉÉ > > ³ ³((��‡‡üüqqææ[ ["Ð"Ð$E%E'º'º)/*/,¤,¤./1Ž1Ž346x6x8í8í:b;b=×=×?L@LBÁBÁD6E6G«G«I J L•L•N O QQSôSôUiViXÞXÞZS[S]È]È_=`=b²b²d'e'gœgœijl†l†nønøpjqjsÜsÜuNvNxÀxÀz2{2}¤}¤€‚ˆ‚ˆ„ú„ú†l‡l‰Þ‰Þ‹PŒPŽÂŽÂ�4‘4“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ñ}!1�Ñ}!1o¾!1 o¾!1¿Â!1¿Â!1£ç!1€£ç!1A("1üA("1‘,"1‘,"1uQ"1puQ"1’"1ì’"1c–"1ôc–"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1)0#1ì)0#1õ~#1L õ~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1A�&1T A�&1S'1˜S'1¹e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�Þ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�Þ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�Þ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1m�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm æ æ__ØØ)Q*Q4Ê4Ê>A?AI¸I¸S-T-^ ^ his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmææ__ØØ Q Q Ê ÊCC¼¼55®®''   ’ ’" # %„%„'ý'ý)v*v,ï,ï.h/h1á1á3Z4Z6Ó6Ó8L9L;Å;Å=>>>@·@·B0C0E©E©G"H"J™J™LMO‡O‡QþQþSuTuVìVìXcYc[Ú[Ú]Q^Q`È`Èb=c=e²e²g'h'jœjœlmo„o„q÷q÷sjtjvÝvÝxPyP{Ã{Ã}6~6€©€©‚ƒ…�…�‡ˆŠrŠrŒäŒäŽV�V‘ȑȓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1+~!1�+~!1ɾ!1 ɾ!1)Å!1)Å!1ýç!1€ýç!1›("1ü›("1û."1û."1ÏQ"1pÏQ"1m’"1ìm’"1͘"1ô͘"1 Ù"1` Ù"1ký"1Ìký"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µk7 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1zº€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µk7 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1zº€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µk7 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1zº €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Íí  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ä ä^^ÔÔ)J*J4À4À>4?4I¨I¨ST^�^�hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjää^^ØØ R R Ì ÌFFÀÀ::´´..¨¨"" œ œ"#%Œ%Œ'(*x*x,î,î.d/d1Ú1Ú3P4P6Æ6Æ8<9<;²;²=(>(@ž@žBCEŠEŠGHJtJtLèLèN\O\QÐQÐSDTDV¸V¸X,Y,[ [ ]^`ˆ`ˆbübüdpepgägäiXjXlÌlÌn@o@q´q´s(t(vœvœxy{„{„}ø}øl€l‚à‚à„T…T‡ȇȉ<Š<Œ°Œ°Ž$�$‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ñ}!1�Ñ}!1o¾!1 o¾!1¿Â!1¿Â!1�ç!1€�ç!1;("1ü;("1‘,"1‘,"1oQ"1poQ"1 ’"1ì ’"1c–"1ôc–"1»Ú"1`»Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›Ú b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›Ú b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›Ú b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1~ž  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk â âYYÐÐ)G*G4¾4¾>3?3I¨I¨ST^Œ^Œhýhýrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkââYYÐÐ G G ¾ ¾55¬¬##ššˆˆÿÿv v"í"í$d%d'Û'Û)R*R,É,É.@/@1·1·3.4.6¥6¥89;“;“= > @�@�BøBøDoEoGæGæI[J[LÐLÐNEOEQºQºS/T/V¤V¤XY[Ž[Ž]^`x`xbëbëd^e^gÑgÑiDjDl·l·n(o(q™q™s t v{v{xìxìz]{]}Î}Î?€?‚°‚°„!…!‡‘‡‘‰ŠŒsŒsŽäŽä�U‘U“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1+~!1�+~!1ɾ!1 ɾ!1)Å!1)Å!1ýç!1€ýç!1›("1ü›("1û."1û."1ÏQ"1pÏQ"1m’"1ìm’"1͘"1ô͘"1 Ù"1` Ù"1ký"1Ìký"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1™€&1T ™€&1iR'1˜iR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µm9 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1zº€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µm9 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1zº€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µm9 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1zº €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âZZÎÎ)B*B4¶4¶>(?(IšIšS T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢’’  ‚ ‚"ú"ú$n%n'â'â)V*V,Ê,Ê.>/>1²1²3&4&6š6š89;‚;‚=ö=ö?j@jBÞBÞDRERGÆGÆI8J8LªLªNOQŽQŽSTVrVrXäXäZV[V]È]È_:`:b¬b¬deg�g�ijltltnænæpXqXsÊsÊuŒ>ްް�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ë}!1�Ë}!1i¾!1 i¾!1­Â!1­Â!1�ç!1€�ç!1;("1ü;("1,"1,"1oQ"1poQ"1 ’"1ì ’"1Q–"1ôQ–"1£Ú"1`£Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1G�&1T G�&1S'1˜S'1¿e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÛ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÛ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÛ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1hˆ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ii á áYYÑÑ)I*I4Á4Á>7?7I­I­S#T#^™^™his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðiiááYYÑÑ I I Á Á99±±))¡¡‘‘  � �"ù"ù$q%q'é'é)a*a,Ù,Ù.Q/Q1É1É3A4A6¹6¹8191;©;©=!>!@™@™BCE‰E‰GHJwJwLíLíNcOcQÙQÙSOTOVÅVÅX;Y;[±[±]'^'`�`�bce‰e‰gÿgÿiujulëlënaoaq×q×sMtMvÃvÃx9y9{¯{¯}%~%€›€›‚ƒ…‡…‡‡ý‡ý‰sŠsŒéŒéŽ_�_‘Õ‘Õ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!11~!1�1~!1Ͼ!1 Ͼ!1MÅ!1MÅ!1è!1€è!1¡("1ü¡("1/"1/"1ÕQ"1pÕQ"1s’"1ìs’"1ñ˜"1ôñ˜"1‹Ú"1`‹Ú"1ëþ"1Ìëþ"1)0#1ì)0#1õ~#1L õ~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1‡€&1T ‡€&1WR'1˜WR'1ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оt@ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¢â€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оt@ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¢â€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оt@ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¢â €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Oo  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84«4«>?I‘I‘ST^w^whêhêr[s[} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=×=×?J@JB½B½D0E0G£G£IJL‰L‰NüNüPoQoSâSâUUVUXÈXÈZ;[;]®]®_!`!b”b”degzgziíiík`l`nÓnÓpFqFs¹s¹u,v,xŸxŸz{}…}…øø�k‚k„Þ„Þ†O‡O‰Á‰Á‹3Œ3ޥޥ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Å}!1�Å}!1c¾!1 c¾!1­Â!1­Â!1—ç!1€—ç!15("1ü5("1,"1,"1iQ"1piQ"1’"1ì’"1Q–"1ôQ–"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Ï€&1T Ï€&1ŸR'1˜ŸR'1Ge'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜ × b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜ × b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜ × b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1‹«  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg Ý ÝSSÉÉ)?*?4µ4µ>)?)I�I�ST^…^…hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÝÝSSÉÉ ? ? µ µ++¡¡��yyïïe e"Û"Û$Q%Q'Ç'Ç)=*=,³,³.)/)1Ÿ1Ÿ346‹6‹89;w;w=í=í?c@cBÙBÙDOEOGÅGÅI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒuŒuŽéŽé�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!11~!1�1~!1Ͼ!1 Ͼ!1MÅ!1MÅ!1è!1€è!1¡("1ü¡("1/"1/"1ÕQ"1pÕQ"1s’"1ìs’"1ñ˜"1ôñ˜"1£Ú"1`£Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1¥$1¬ ¥$1}ò$1p }ò$1·Þ%1t ·Þ%1'€&1T '€&1÷Q'1˜÷Q'1Ÿd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оvB b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1£ã€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оvB b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1£ã€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1оvB b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1£ã €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1hˆ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛNNÁÁ)4*44¥4¥>?I‡I‡SøSø]i^ihÚhÚrIsI} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹þþqqääW W"Ê"Ê$=%='°'°)#*#,–,–. / 1|1|3ï3ï5b6b8Õ8Õ:H;H=¹=¹?*@*B›B›D E G}G}IîIîK_L_NÐNÐPAQAS²S²U#V#X”X”Z[]v]v_ç_çaXbXdÉdÉf:g:i«i«kln�n�pþpþrosouàuàwQxQzÂzÂ|3}3¤¤�‚„†„††õ†õˆe‰e‹Õ‹Õ�EŽE�µ�µ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Å}!1�Å}!1c¾!1 c¾!1³Â!1³Â!1—ç!1€—ç!15("1ü5("1…,"1…,"1iQ"1piQ"1’"1ì’"1W–"1ôW–"1—Ú"1`—Ú"1÷þ"1Ì÷þ"1#0#1ì#0#1ï~#1L ï~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c— Ø b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c— Ø b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c— Ø b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1cƒ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>-?-I¢I¢ST^Ž^Žhisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbçbçd\e\gÑgÑiFjFl»l»n2o2q§q§stv‘v‘xy{{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!17~!1�7~!1Õ¾!1 Õ¾!1#Å!1#Å!1è!1€è!1¡("1ü¡("1õ."1õ."1ÕQ"1pÕQ"1s’"1ìs’"1ǘ"1ôǘ"1‹Ú"1`‹Ú"1ëþ"1Ìëþ"1#0#1ì#0#1ï~#1L ï~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Õ€&1T Õ€&1¥R'1˜¥R'1Me'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Ç}I b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1y¹€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Ç}I b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1y¹€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“Ç}I b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1y¹ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Vv  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ú ÚOOÄÄ)9*94®4®>"?"I–I–ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÚÚRRÊÊ B B º º22ªª""šš„„ùùn n"ã"ã$X%X'Í'Í)B*B,·,·.,/,1¡1¡346‹6‹89;u;u=ê=ê?_@_BÔBÔDIEIG¾G¾I2J2L¦L¦NOQŽQŽSTVvVvXêXêZ^[^]Ò]Ò_F`FbÄbÄd8e8g¬g¬i j l”l”noq|q|sðsðudvdxØxØzL{L}À}À4€4‚¨‚¨„…‡�‡�‰ŠŒzŒzŽïŽï�d‘d“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1¿}!1�¿}!1]¾!1 ]¾!1³Â!1³Â!1‘ç!1€‘ç!1/("1ü/("1…,"1…,"1cQ"1pcQ"1’"1ì’"1W–"1ôW–"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ô b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ô b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ô b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ž®  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÅÅ):*:4¯4¯>"?"I•I•ST^^hòhòrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=ß=ß?T@TBÉBÉD>E>G³G³I&J&L™L™N O QQSòSòUeVeXØXØZK[K]¾]¾_1`1b¤b¤degŠgŠiýiýkplpnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!17~!1�7~!1Õ¾!1 Õ¾!1#Å!1#Å!1 è!1€ è!1§("1ü§("1õ."1õ."1ÛQ"1pÛQ"1y’"1ìy’"1ǘ"1ôǘ"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1#0#1ì#0#1ï~#1L ï~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“ÇK b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1y¹€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“ÇK b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1y¹€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1“ÇK b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1y¹ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1s“  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØMMÂÂ)7*74¬4¬>?I’I’S T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?M@MBÂBÂD7E7G¬G¬IJL’L’NOQxQxSëSëU^V^XÑXÑZD[D]·]·_*`*b¤b¤degˆgˆiúiúklllnÞnÞpPqPsÂsÂu4v4x¦x¦z{}Š}Šüü�n‚n„à„à†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L7i!1¿}!1�¿}!1]¾!1 ]¾!1¡Â!1¡Â!1‹ç!1€‹ç!1)("1ü)("1s,"1s,"1]Q"1p]Q"1û‘"1ìû‘"1E–"1ôE–"1=Ú"1`=Ú"1�þ"1Ô�þ"10#1ô0#1é~#1T é~#1Ï$1´ Ï$1÷ð$1x ÷ð$1ÕÀ%1| ÕÀ%1ç€&1\ ç€&1ÝP'1 ÝP'1…c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” Õ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” Õ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” Õ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1¿OÁ-  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee ß ßUUÍÍ)C*C4·4·>+?+IŸIŸST^“^“hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯%%›› ‡ ‡"ý"ý$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@ƒ@ƒB÷B÷DkEkGßGßISJSLÇLÇN;O;Q¯Q¯S#T#V—V—X Y [[]ó]ó_g`gbçbçd[e[gÏgÏiCjCl·l·n+o+qŸqŸstv‡v‡xûxûzo{o}ã}ãW€W‚˂˄?…?‡±‡±‰$Š$Œ—Œ—Ž � ‘}‘}“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!1=~!1�=~!1Û¾!1 Û¾!1AÅ!1AÅ!1 è!1€ è!1§("1ü§("1/"1/"1ÛQ"1pÛQ"1y’"1ìy’"1å˜"1ôå˜"1ÛØ"1`ÛØ"1;ý"1Ì;ý"10#1ì0#1é~#1L é~#1Õ$1¬ Õ$1ýð$1p ýð$1ÛÀ%1t ÛÀ%1ÿ€&1T ÿ€&1õP'1˜õP'1�c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐ†R b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 à€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐ†R b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 à€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐ†R b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 à €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ÏË  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ Õ ÕLLÍÍ)D*D4»4»>2?2I©I©S T ^ž^žhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÕÕLLÃà : : ± ±((ŸŸ��{{òòi i"à"à$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÑLÑNHOHQ¿Q¿S6T6V­V­X$Y$[›[›]^`‰`‰bcewewgîgîiejelÜlÜnZoZqÌqÌs>t>v°v°x"y"{”{”}~€x€x‚ê‚ê„\…\‡·Ή@Š@Œ²Œ²Ž$�$‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!1¹}!1�¹}!1W¾!1 W¾!1¡Â!1¡Â!1‹ç!1€‹ç!1)("1ü)("1s,"1s,"1]Q"1p]Q"1û‘"1ìû‘"1E–"1ôE–"1=Ú"1`=Ú"1�þ"1Ô�þ"10#1ô0#1é~#1T é~#1·$1´ ·$1ßð$1x ßð$1½À%1| ½À%1o€&1\ o€&1eP'1 eP'1 c'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ñ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ñ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1^’Ñ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1Ï_¼(  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛOOÇÇ);*;4­4­>?I‘I‘ST^�^�hóhórcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››ƒƒ÷÷k k"ß"ß$W%W'Ë'Ë)?*?,³,³.'/'1›1›346ƒ6ƒ8÷8÷:k;k=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b—b—d e g{g{iíiík_l_nÑnÑpCqCsµsµu'v'x™x™z { }}}}ïï�a‚a„Ó„Ó†C‡C‰´‰´‹%Œ%Ž–Ž–�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L7i!1=~!1�=~!1Û¾!1 Û¾!1GÅ!1GÅ!1è!1€è!1­("1ü­("1/"1/"1áQ"1páQ"1’"1ì’"1ë˜"1ôë˜"1ÛØ"1`ÛØ"1;ý"1Ì;ý"10#1ì0#1é~#1L é~#1Ã$1¬ Ã$1ëð$1p ëð$1ÉÀ%1t ÉÀ%1¥€&1T ¥€&1›P'1˜›P'1Cc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐˆT b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¡á€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐˆT b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¡á€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1œÐˆT b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¡á €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ÍÉ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10^^ Ó ÓHHÅÅ):*:4¯4¯>$?$I™I™ST^Œ^Œhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð^^ÓÓHH½½ 2 2 § §‘‘{{ððeeÚÚO O"Ä"Ä$A%A'¶'¶)+*+, , ./1Š1Š3ÿ3ÿ5t6t8é8é:^;^=Ó=Ó?H@HB½B½D2E2G§G§IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚noqrqrsäsäuVvVxÈxÈz:{:}¬}¬€‚�‚�„…‡t‡t‰æ‰æ‹XŒXŽÊŽÊ�<‘<“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„1i!1q}!1q}!1¾!1|¾!1sÃ!1„sÃ!1©Æ!1ð©Æ!13Ð!1|3Ð!1iÓ!1HiÓ!1=ç!1Ô=ç!1Û'"1PÛ'"1E-"1XE-"1{0"1Ä{0"1:"1P:"1;="1;="1Q"1¨Q"1­‘"1$ ­‘"1—"1, —"1Mš"1˜ Mš"1×£"1$ ×£"1 §"1ð  §"1OÔ"1| OÔ"1=Ú"1X=Ú"1�þ"1Ü�þ"1ß+#1üß+#1�:#1Ø�:#1Yo#1dYo#1]�#1ˆ]�#1[š#1´[š#1É$1@É$1¡ò$1¡ò$1ÛÞ%1ÛÞ%1Uì%1èUì%1Õ€&1tÕ€&1¥R'1¸¥R'1Me'1¼€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PÇ“ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ï bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PÇ“ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ï  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PÇ“  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ï bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïï‚¢4 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1_Ÿþ>Z�Ý<|  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d_ß>¾�ü|Û[º:™xø\ÜW×'§6¶òr • ½= ô t Ó S ² 2 ‘  p ð Ê O�î.�Í,l bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ × ×  ËË)@*@4µ4µ>+?+IŸIŸST^“^“his}s}} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__××OOÇÇ ? ? · ·//§§——ÊÊ??´´) )"ž"ž$_%_'Ô'Ô)I*I,¾,¾.3/31¨1¨346’6’89;|;|=ñ=ñ?f@fBÛBÛDPEPGÅGÅI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÏbÏdCeCg·g·i+j+lŸlŸnoq‡q‡sûsûuovoxãxãzW{W}Ë}Ë?€?‚³‚³„'…'‡�‡�‰ŠŒ‡Œ‡ŽüŽü�q‘q“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1O~!1èO~!1í¾!1dí¾!1¡Å!1l¡Å!1+Ï!1è+Ï!1aÒ!1´aÒ!1!è!1@!è!1¿("1¼¿("1s/"1Äs/"1ý8"1@ý8"13<"1 3<"1óQ"1˜óQ"1‘’"1‘’"1E™"1 E™"1Ï¢"1˜ Ï¢"1¦"1d ¦"1GÓ"1ð GÓ"1ÇÚ"1Ì ÇÚ"1'ÿ"1H 'ÿ"1i,#1hi,#1�:#1D�:#1Yo#1ÐYo#1]�#1ô]�#1[š#1 [š#1Ï$1¬Ï$1§ò$1p§ò$1áÞ%1táÞ%1[ì%1T[ì%1�&1à�&1ÕR'1$ÕR'1}e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¸ì¤p b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _Ÿa¡þ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¸ì¤p b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _Ÿa¡þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¸ì¤p  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _Ÿa¡þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 |œ(œ¼ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1_Ÿþ>N�Ý<|  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d_ß>¾�ü|Û[º:™xø9¹W×þ~ÃC6¶Š  • ô t Ó S ² 2 “  r ò s ³ Q‘ð0�Ï.n bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPP‡‡)@*@4·4·>.?.I£I£ST^™^™hisƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$%'’'’) * ,€,€.÷.÷0°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbèbèd]e]gÒgÒiGjGl¼l¼n=o=q²q²s't'vœvœxy{†{†}û}ûp€p‚å‚å„Z…Z‡χωDŠDŒ¹Œ¹Ž.�.‘£‘£“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„1i!1k}!1k}!1 ¾!1| ¾!1sÃ!1„sÃ!1©Æ!1ð©Æ!13Ð!1|3Ð!1iÓ!1HiÓ!1=ç!1Ô=ç!1Û'"1PÛ'"1E-"1XE-"1{0"1Ä{0"1:"1P:"1;="1;="1Q"1¨Q"1­‘"1$ ­‘"1—"1, —"1Mš"1˜ Mš"1×£"1$ ×£"1 §"1ð  §"1OÔ"1| OÔ"1ãÙ"1XãÙ"1Cþ"1ÜCþ"1…+#1ü…+#1÷:#1Ø÷:#1³o#1d³o#1‘#1ˆ‘#1œ#1´œ#1½$1@½$1•ò$1•ò$1ÏÞ%1ÏÞ%1Iì%1èIì%1�€&1t�€&1]R'1¸]R'1e'1¼€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OÄ� b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ï bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OÄ� b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ï  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OÄ�  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ï bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïïkgg?qÝ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1v¶,là ¥å ”Ô  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dvölìbâXØNÎDÄ:º0°$¤ ˜ ˆ   Œ S Ó  €  ž ô t é i è h Ü \U•ÐPÆ|¼0pä$ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Õ ÕÇÇ)<*<4±4±>&?&I™I™ST^‰^‰hûhûroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……··,,¡¡ "‹"‹$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ºî¨t b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _Ÿb¢þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ºî¨t  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _Ÿb¢þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 k‹&m� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1_Ÿþ>�Ý“Ó <|  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d_ß>¾�ü|Û[º:™xøW×6¶ • O Ï ô t  ” Ó S Ù Y ² 2 ž  ‘  p ð O�ø 8î.�Í,l bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×MMƒƒ);*;4°4°>%?%I˜I˜S T ^Š^Šhýhýrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$ % '€'€)õ)õ+j,j.ß.ß0—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnònòpeqesØsØuKvKx¾x¾z1{1}¤}¤€‚ЂЄý„ý†p‡p‰ã‰ã‹VŒVŽÉŽÉ�<‘<“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx7i!1¹}!1è¹}!1W¾!1dW¾!1§Å!1l§Å!11Ï!1è1Ï!1gÒ!1´gÒ!1‹ç!1@‹ç!1)("1¼)("1y/"1Äy/"19"1@9"19<"1 9<"1]Q"1˜]Q"1û‘"1û‘"1K™"1 K™"1Õ¢"1˜ Õ¢"1 ¦"1d  ¦"1MÓ"1ð MÓ"1CÚ"1Ì CÚ"1£þ"1P £þ"1å+#1på+#1£:#1L£:#1_o#1Ø_o#1]�#1ü]�#1[š#1([š#1É$1´É$1¡ò$1x¡ò$1ÛÞ%1|ÛÞ%1Uì%1\Uì%1ó€&1èó€&1ÃR'1,ÃR'1ke'10€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“ Ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @` ÿ? bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“ Ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @` ÿ? bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“ Ö  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @` ÿ?  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ï6 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1_Ÿþ>U�Ý<|  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d_ß>¾�ü|Û[º:™xøKËW×”6¶Ý]¦&  • ô t Ó S ² 2 ‘  p ð ¥ å O�î.�Í,l bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Ö ÖMMÄÄ);*;4²4²>+?+I¢I¢S×S×]•^•hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÖÖMMÄÄ ; ; ² ²))  ŽŽ||óój j"á"á$X%X'Ï'Ï)F*F,½,½.4/41«1«3"4"6™6™89;‡;‡=þ=þ?u@uBìBìDcEcGÚGÚISJSLÊLÊNAOAQ¸Q¸S/T/V¦V¦XY[”[”] ^ `‚`‚b·b·d.e.g¥g¥ijl“l“nQoQqÃqÃs5t5v§v§xy{‹{‹}ý}ýo€o‚á‚á„S…S‡Ňʼn7Š7Œ©Œ©Ž�‘�‘�“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx7i!1I~!1èI~!1ç¾!1dç¾!1¡Å!1l¡Å!1+Ï!1è+Ï!1aÒ!1´aÒ!1è!1@è!1¹("1¼¹("1s/"1Äs/"1ý8"1@ý8"13<"1 3<"1íQ"1˜íQ"1‹’"1‹’"1E™"1 E™"1Ï¢"1˜ Ï¢"1¦"1d ¦"1GÓ"1ð GÓ"1Ú"1Ì Ú"1yþ"1H yþ"1»+#1h»+#1Í:#1DÍ:#1‰o#1Љo#1;�#1ô;�#19›#1 9›#1Ã$1¬Ã$1›ò$1p›ò$1ÕÞ%1tÕÞ%1Oì%1TOì%1Û€&1àÛ€&1«R'1$«R'1Se'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1­á™e b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _Ÿ±ñþ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1­á™e b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _Ÿ±ñþ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1­á™e  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _Ÿ±ñþ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 o�Ïïî bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1i©R$d½ýh¨  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1diéRÒ;»$¤ �övß_ÈH�³3`à0°ž € ‰ t ô _ ß J Ê 5 µ   4t I²ò]�H bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff à àVVÌÌ)B*B4¸4¸>,?,I I SÔSÔ]Š^Šhþhþrrsr} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffààZZÔÔ N N È ÈBB¼¼66°°&&œœ ˆ ˆ"þ"þ$t%t'ê'ê)`*`,Ö,Ö.L/L1Â1Â38486®6®8$9$;š;š=>@„@„BøBøDlElGàGàITJTLÈLÈN�ÝžÞ <|  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d_ß>¾�ü|Û[º:™xøW×6¶ • o ï ô t 8 ¸ Ó S  � ² 2 Ê J ‘  p ð O�.nî.�Í,l bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Ô ÔII¾¾)3*34¨4¨>?I”I”SÆSÆ]ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÔÔII¾¾ 3 3 ¨ ¨’’||ññffÛÛP P"Å"Å$:%:'¯'¯)$*$,™,™./1ƒ1ƒ3ø3ø5m6m8â8â:W;W=Ì=Ì?A@AB¶B¶D+E+G G IJLŒLŒNOQvQvSëSëU`V`XÕXÕZJ[J]¿]¿_4`4bfbfdÛdÛfPgPiÅiÅk:l:n÷n÷piqisÛsÛuMvMx¿x¿z1{1}£}£€‚‡‚‡„ù„ù†k‡k‰݉݋OŒOŽÁŽÁ�3‘3“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx7i!1O~!1èO~!1í¾!1dí¾!1¡Å!1l¡Å!1+Ï!1è+Ï!1aÒ!1´aÒ!1!è!1@!è!1¿("1¼¿("1s/"1Äs/"1ý8"1@ý8"13<"1 3<"1óQ"1˜óQ"1‘’"1‘’"1E™"1 E™"1Ï¢"1˜ Ï¢"1¦"1d ¦"1GÓ"1ð GÓ"1%Ú"1Ì %Ú"1…þ"1H …þ"1Ç+#1hÇ+#1£:#1D£:#1_o#1Ð_o#1]�#1ô]�#1[š#1 [š#1«$1¬«$1ƒò$1pƒò$1½Þ%1t½Þ%17ì%1T7ì%1i€&1ài€&19R'1$9R'1ád'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¯ã�i b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _Ÿ°ðþ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¯ã�i b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _Ÿ°ðþ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¯ã�i  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _Ÿ°ðþ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 f†Ïïý bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1_Ÿþ>�Ý´ô <|  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d_ß>¾�ü|Û[º:™xøW×6¶ • ô t Ð P Ó S   ² 2 p ð ‘  @ À p ð O�î.Ä�Í,l bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜPPÄÄ)8*84¬4¬>?I�I�SÃSÃ]x^xhêhêr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœ„„øøl l"à"à$T%T'È'È)<*<,°,°.$/$1˜1˜3 4 6€6€8ô8ô:h;h=Ü=Ü?N@NBÀBÀD2E2G¤G¤IJLˆLˆNúNúPlQlSÞSÞUPVPXÂXÂZ4[4]¦]¦_`bKbKd½d½f/g/i¡i¡klnÈnÈp:q:s¬s¬uvx�x�z{}t}tææ�X‚X„ʄʆ<‡<‰­‰­‹ŒŽ�Ž��‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¿}!1п}!1]¾!1L]¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1‘ç!1(‘ç!1/("1¤/("1y/"1¬y/"19"1(9"19<"1ô9<"1cQ"1€cQ"1’"1ü’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"11Ú"1´ 1Ú"1‘þ"18 ‘þ"1Ó+#1XÓ+#1s7#14s7#1±Œ#1 ±Œ#1É$1,É$1¡ò$1ð¡ò$1ÛÞ%1ôÛÞ%1ÿ€&1Ôÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” × b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U_Ÿþ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” × b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 U_Ÿþ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” ×  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 U_Ÿþ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïïzš$ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1N€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations18¸ÿÆF� ~ ¾  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ý ÝWWÍÍ)C*C4¹4¹>/?/I£I£ST^‹^‹hÿhÿrsss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÝÝWWÑÑ K K Å Å??¹¹33­­''¡¡ • •"#%…%…'û'û)q*q,ç,ç.]/]1Ó1Ó3I4I6¿6¿8595;«;«=!>!@—@—B C EƒEƒGùGùIoJoLãLãNWOWQËQËS?T?V³V³X'Y'[›[›]^`ƒ`ƒb÷b÷dkekgßgßiSjSlÇlÇn;o;q¯q¯s#t#v—v—x y {{}ó}óg€g‚Û‚Û„O…O‡ÇÉ7Š7Œ«Œ«Ž�‘“‘““ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1C~!1ÐC~!1á¾!1Lá¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1è!1(è!1³("1¤³("1/"1¬/"1 9"1( 9"1?<"1ô?<"1çQ"1€çQ"1…’"1ü…’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1Ú"1´ Ú"1aþ"10 aþ"1£+#1P£+#1£7#1,£7#1+Ž#1˜+Ž#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1/�&1Ì/�&1ÿR'1ÿR'1§e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖŽZ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _Ÿw·B bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖŽZ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _Ÿw·B bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¢ÖŽZ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _Ÿw·B  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 b‚ÏïÔô bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1:z€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1èhÃCž y ù ú: €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff ß ßXXÑÑ)J*J4Ã4Ã>7@°@°B)C)E¢E¢GHJ”J”L M O‚O‚QùQùSpTpVçVçX^Y^[Õ[Õ]L^L`Ã`Ãb8c8e­e­g"h"j—j—l m ooqòqòsetevØvØxKyK{¾{¾}1~1€¤€¤‚ƒ…ЅЇý‡ý‰oŠoŒáŒáŽS�S‘őœ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1¿}!1п}!1]¾!1L]¾!1§Å!1T§Å!11Ï!1Ð1Ï!1gÒ!1œgÒ!1‘ç!1(‘ç!1/("1¤/("1y/"1¬y/"19"1(9"19<"1ô9<"1cQ"1€cQ"1’"1ü’"1K™"1 K™"1Õ¢"1€ Õ¢"1 ¦"1L  ¦"1MÓ"1Ø MÓ"1ÝÙ"1´ ÝÙ"1=þ"18 =þ"1+#1X+#1û9#14û9#15�#1 5�#1±$1,±$1‰ò$1ð‰ò$1ÃÞ%1ôÃÞ%1‡€&1Ô‡€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” Õ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 V_Ÿþ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” Õ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 V_Ÿþ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” Õ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 V_Ÿþ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïïgcc;cÏ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1•Õ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1T Ô  › â b © ) E €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛSSÇÇ);*;4¯4¯>#?#I•I•ST^y^yhëhër]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››‹‹ { {"ó"ó$g%g'Û'Û)O*O,Ã,Ã.7/71«1«346“6“89;{;{=ï=ï?c@cB×B×DKEKG¿G¿I3J3L¥L¥NOQ‰Q‰SûSûUmVmXßXßZQ[Q]Ã]Ã_5`5b§b§deg‹g‹iýiýkolonánápSqSsÅsÅu7v7x©x©z{}�}�ÿÿ�q‚q„ã„ã†U‡U‰ljNj9Œ9ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1I~!1ÐI~!1ç¾!1Lç¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1è!1(è!1³("1¤³("1/"1¬/"1 9"1( 9"1?<"1ô?<"1çQ"1€çQ"1…’"1ü…’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1Ú"1´ Ú"1yþ"10 yþ"1»+#1P»+#1=:#1,=:#1ÇŽ#1˜ÇŽ#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1½€&1̽€&1�R'1�R'15e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¤Ø’^ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _Ÿw·C bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¤Ø’^ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _Ÿw·C bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¤Ø’^  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _Ÿw·C  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ]}Ïïê  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Õ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1T Ô / ¯ Š å e •Õ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>.?.I£I£ST^‡^‡høhørjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbæbædYeYgÌgÌi?j?l²l²n#o#q”q”stvvvvxçxçzX{X}É}É:€:‚«‚«„…‡އމÿ‰ÿ‹pŒpŽáŽá�R‘R“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1Å}!1ÐÅ}!1c¾!1Lc¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1—ç!1(—ç!15("1¤5("1/"1¬/"1 9"1( 9"1?<"1ô?<"1iQ"1€iQ"1’"1ü’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1�Ú"1´ �Ú"1ýþ"18 ýþ"1?,#1X?,#1‹7#14‹7#1e�#1 e�#1·$1,·$1�ò$1ð�ò$1ÉÞ%1ôÉÞ%1‡€&1Ô‡€&1WR'1WR'1ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a• Ø b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @a¡@ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a• Ø b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @a¡@ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a• Ø  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @a¡@  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1o�îêd„ê bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1 `€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€RÒ$¤öv 0p €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10`` Ö ÖKKÀÀ)5*54ª4ª>?I�I�ST^v^vhéhér\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð``ÖÖLL 8 8 ® ®$$šš††ûûppååZ Z"Ï"Ï$D%D'¹'¹).*.,£,£./1�1�346w6w8ì8ì:a;a=Ö=Ö?I@IB¼B¼D/E/G¢G¢IJLˆLˆNûNûPnQnSáSáUTVTXÇXÇZ:[:]­]­_ ` b“b“degygyiìiìk_l_nÒnÒpEqEs¸s¸u+v+xžxžz{}„}„÷÷�j‚j„݄݆P‡P‰‰‹4Œ4ަަ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1=~!1Ð=~!1Û¾!1LÛ¾!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1è!1(è!1­("1¤­("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1áQ"1€áQ"1’"1ü’"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1ïÙ"1´ ïÙ"1Oþ"10 Oþ"1‘+#1P‘+#1»7#1,»7#1÷Ž#1˜÷Ž#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1A�&1ÌA�&1S'1S'1¹e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—˃O b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _Ÿ¦æ M bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—˃O b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _Ÿ¦æ M bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—˃O  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _Ÿ¦æ M  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÏïUuºÚ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1P�€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1@À&¦ Œ ò r À €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÎÎ)F*F4¾4¾>6?6I¬I¬S"T"^˜^˜his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$n%n'æ'æ)^*^,Ö,Ö.N/N1Æ1Æ3>4>6¶6¶8.9.;¦;¦=>@–@–BCE†E†GþGþIvJvLìLìNbObQØQØSNTNVÄVÄX:Y:[°[°]&^&`œ`œbceˆeˆgþgþitjtlêlên`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1Å}!1ÐÅ}!1c¾!1Lc¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1‘ç!1(‘ç!1/("1¤/("1/"1¬/"1 9"1( 9"1?<"1ô?<"1cQ"1€cQ"1’"1ü’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1yÚ"1´ yÚ"1Ùþ"18 Ùþ"1,#1X,#1:#14:#1û�#1 û�#1¥$1,¥$1}ò$1ð}ò$1·Þ%1ô·Þ%1'€&1Ô'€&1÷Q'1÷Q'1Ÿd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a• Ö b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @` ÿ? bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a• Ö b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @` ÿ? bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a• Ö  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @` ÿ?  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1`\\4CcWà bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1²ò €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1È H š  l ì > ¾  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØKK¾¾)1*14¤4¤>?I†I†S÷S÷]h^hhÙhÙrJsJ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆûûnnááT T"Ç"Ç$:%:'­'­) * ,“,“./1y1y3ì3ì5_6_8Ò8Ò:E;E=¸=¸?)@)BšBšD E G|G|IíIíK^L^NÏNÏP@Q@S±S±U"V"X“X“Z[]u]u_æ_æaWbWdÈdÈf9g9iªiªklnŒnŒpýpýrnsnußußwPxPzÁzÁ|2}2££�‚„…„…†ö†öˆf‰f‹Ö‹Ö�FŽF�¶�¶’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1C~!1ÐC~!1á¾!1Lá¾!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1è!1(è!1­("1¤­("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1áQ"1€áQ"1’"1ü’"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1 Ú"1´  Ú"1mþ"10 mþ"1¯+#1P¯+#1a:#1,a:#1Ÿ�#1˜Ÿ�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1É€&1ÌÉ€&1™R'1™R'1Ae'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1™Í‡S b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _Ÿ¥åN bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1™Í‡S b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _Ÿ¥åN bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1™Í‡S  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _Ÿ¥åN  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÏïTtØø bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1ö6 €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Ø X ¾ > ¤ $ Š f¦ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÆÆ)<*<4²4²>(?(IœIœST^„^„høhørlsl} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$N%N'Ä'Ä):*:,°,°.&/&1œ1œ346ˆ6ˆ8þ8þ:t;t=ê=ê?`@`BÖBÖDLELGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡Œ‡Œ‰ŠŒtŒtŽèŽè�\‘\“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1Ë}!1ÐË}!1i¾!1Li¾!1ÅÅ!1TÅÅ!1OÏ!1ÐOÏ!1…Ò!1œ…Ò!1�ç!1(�ç!1;("1¤;("1—/"1¬—/"1!9"1(!9"1W<"1ôW<"1oQ"1€oQ"1 ’"1ü ’"1i™"1 i™"1ó¢"1€ ó¢"1)¦"1L )¦"1kÓ"1Ø kÓ"1»Ú"1´ »Ú"1ÿ"18 ÿ"1],#1X],#1£7#14£7#11Ž#1 1Ž#1É$1,É$1¡ò$1ð¡ò$1ÛÞ%1ôÛÞ%1Õ€&1ÔÕ€&1¥R'1¥R'1Me'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– Ù b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Iw·V bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– Ù b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Iw·V bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– Ù  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Iw·V  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1d„ØÔ�¡Ôø bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations16v€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ØXµ5’ o ï ö6 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Û ÛPPÅÅ):*:4¯4¯>%?%I™I™S T ^�^�hõhõrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÛÛSSËË C C » »33««##››……úúo o"ä"ä$Y%Y'Î'Î)C*C,¸,¸.-/-1¢1¢346Œ6Œ89;v;v=ë=ë?`@`BÕBÕDJEJG¿G¿I5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`Ib½b½d1e1g¥g¥ijl�l�noququséséu]v]xÑxÑzE{E}¹}¹-€-‚¡‚¡„…‡‹‡‹‰ŠŒuŒuŽêŽê�_‘_“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!17~!1Ð7~!1Õ¾!1LÕ¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1è!1(è!1¡("1¤¡("1…/"1¬…/"19"1(9"1E<"1ôE<"1ÕQ"1€ÕQ"1s’"1üs’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1×Ù"1´ ×Ù"17þ"10 7þ"1y+#1Py+#1Ó7#1,Ó7#1Ã�#1˜Ã�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀxD b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _Ÿw·A bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀxD b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _Ÿw·A bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀxD  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _Ÿw·A  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÏïHh À bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1f¦€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1˜‰ z ú k ë †Æ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ý ÝTTËË)B*B4¹4¹>0?0I¥I¥ST^�^�hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÝÝTTËË B B ¹ ¹00§§••  ƒƒúúq q"è"è$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubêbêd_e_gÔgÔiIjIl¾l¾n3o3q¨q¨stv’v’xy{|{|}ñ}ñf€f‚Û‚Û„P…P‡Ňʼn:Š:Œ¯Œ¯Ž$�$‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1Ë}!1ÐË}!1i¾!1Li¾!1ÅÅ!1TÅÅ!1OÏ!1ÐOÏ!1…Ò!1œ…Ò!1—ç!1(—ç!15("1¤5("1—/"1¬—/"1!9"1(!9"1W<"1ôW<"1iQ"1€iQ"1’"1ü’"1i™"1 i™"1ó¢"1€ ó¢"1)¦"1L )¦"1kÓ"1Ø kÓ"1£Ú"1´ £Ú"1ÿ"18 ÿ"1E,#1XE,#1C:#14C:#1ÍŽ#1 ÍŽ#1½$1,½$1•ò$1ð•ò$1ÏÞ%1ôÏÞ%1�€&1Ô�€&1]R'1]R'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– × b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jv¶U bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– × b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jv¶U bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– ×  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jv¶U  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1YUU-hˆS¿ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ó €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1L Ì ) ©  † ã c “Ó €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ù ÙNNÃÃ)8*84­4­>"?"I•I•ST^y^yhëhër_s_} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÙÙOOÅÅ ; ; ± ±''��‰‰þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?N@NBÃBÃD8E8G­G­I"J"L•L•NOQ{Q{SîSîUaVaXÔXÔZG[G]º]º_-`-bŸbŸdegƒgƒiõiõkglgnÙnÙpKqKs½s½u/v/x¡x¡z{}…}…÷÷�i‚i„Û„Û†O‡O‰‰‹5Œ5ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!17~!1Ð7~!1Õ¾!1LÕ¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1 è!1( è!1§("1¤§("1…/"1¬…/"19"1(9"1E<"1ôE<"1ÛQ"1€ÛQ"1y’"1üy’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1ûÙ"1´ ûÙ"1[þ"10 [þ"1�+#1P�+#1…:#1,…:#1q�#1˜q�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1¥€&1Ì¥€&1uR'1uR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂ|H b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _Ÿw·B bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂ|H b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _Ÿw·B bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂ|H  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _Ÿw·B  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÏïKkÆæ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1W €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1\ Ü M Í > ¾ /¯7w €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÆÆ);*;4°4°>%?%I˜I˜S T ^~^~hñhñrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnænæpYqYsÌsÌu?v?x²x²z%{%}˜}˜ € ‚~‚~„ñ„ñ†d‡d‰׉׋JŒJ޽޽�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1Ñ}!1ÐÑ}!1o¾!1Lo¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1�ç!1(�ç!1;("1¤;("1…/"1¬…/"19"1(9"1E<"1ôE<"1oQ"1€oQ"1 ’"1ü ’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1ÇÚ"1´ ÇÚ"1'ÿ"18 'ÿ"1i,#1Xi,#1Á7#14Á7#1ýŽ#1 ýŽ#1Õ$1,Õ$1­ò$1ð­ò$1çÞ%1ôçÞ%1ÿ€&1Ôÿ€&1ÏR'1ÏR'1we'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ú b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @` ÿ? bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ú b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @` ÿ? bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ú  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @` ÿ?  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1Yy¾†¦¾â bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1LŒ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations10°˜ € è h ¼ü €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>/?/I¦I¦ST^�^�hissss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIWJWLÎLÎNEOEQ¼Q¼S3T3VªVªX!Y![˜[˜]^`†`†býbýdtetgëgëibjblÙlÙnKoKq½q½s/t/v¡v¡xy{…{…}÷}÷i€i‚Û‚Û„M…M‡¿‡¿‰1Š1Œ£Œ£Ž�‘‡‘‡“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!11~!1Ð1~!1Ͼ!1LϾ!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1ýç!1(ýç!1›("1¤›("1…/"1¬…/"19"1(9"1E<"1ôE<"1ÏQ"1€ÏQ"1m’"1üm’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1ÇÚ"1´ ÇÚ"1'ÿ"10 'ÿ"1i,#1Pi,#1ñ7#1,ñ7#1‰�#1˜‰�#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1í€&1Ìí€&1½R'1½R'1ee'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µm9 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _ŸŸßÿ? bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µm9 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _ŸŸßÿ? bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�µm9  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _ŸŸßÿ?  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ;[†¦…¥ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1|¼€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ðpìl è h ä d LŒ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ä äZZÐÐ)F*F4¼4¼>0?0I¤I¤ST^Œ^Œhistst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjää^^ØØ R R Ì ÌFFÀÀ::´´**   Œ Œ"#%x%x'î'î)d*d,Ú,Ú.P/P1Æ1Æ3<4<6²6²8(9(;ž;ž=>@ˆ@ˆBüBüDpEpGäGäIXJXLÌLÌN@O@Q´Q´S(T(VœVœXY[„[„]ø]ø_l`lbàbàdTeTgÈgÈi#?#I˜I˜S T ^^hñhñrcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccØØMM 7 7 ¬ ¬!!––  €€õõjjßßT T"É"É$>%>'³'³)(*(,�,�./1‡1‡3ü3ü5q6q8æ8æ:[;[=Ð=Ð?E@EBºBºD/E/G¤G¤IJL�L�NOQzQzSïSïUdVdXÙXÙZN[N]Ã]Ã_8`8b­b­d"e"g—g—i j l�l�nónópeqes×s×uIvIx»x»z-{-}Ÿ}Ÿ€‚ƒ‚ƒ„õ„õ†g‡g‰ىًKŒK޽޽�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!11~!1Ð1~!1Ͼ!1LϾ!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1è!1(è!1¡("1¤¡("1…/"1¬…/"19"1(9"1E<"1ôE<"1ÕQ"1€ÕQ"1s’"1üs’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1¯Ú"1´ ¯Ú"1ÿ"10 ÿ"1Q,#1PQ,#1©:#1,©:#1I‘#1˜I‘#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1u€&1Ìu€&1ER'1ER'1íd'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·q= b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _ŸŸßÿ? bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·q= b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _ŸŸßÿ? bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·q=  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _ŸŸßÿ?  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 Bb´Ôp� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations18x €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1à ` Ü \ Ø XÔTH €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àTTÈÈ)<*<4°4°>"?"I”I”ST^x^xhêhêr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  ˆˆüüp p"ä"ä$X%X'Ì'Ì)@*@,´,´.(/(1œ1œ346„6„8ø8ø:l;l=à=à?R@RBÄBÄD6E6G¨G¨IJLŒLŒNþNþPpQpSâSâUTVTXÆXÆZ8[8]ª]ª_`bŽbŽdegrgriäiäkVlVnÈnÈp:q:s¬s¬uvx�x�z{}t}tææ�X‚X„ʄʆ<‡<‰­‰­‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1×}!1Ð×}!1u¾!1Lu¾!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1£ç!1(£ç!1A("1¤A("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1uQ"1€uQ"1’"1ü’"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1ÁÚ"1´ ÁÚ"1!ÿ"18 !ÿ"1c,#1Xc,#1Ù7#14Ù7#1Ã�#1 Ã�#1Û$1,Û$1³ò$1ð³ò$1íÞ%1ôíÞ%1�&1Ô�&1áR'1áR'1‰e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Û b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @` ÿ? bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Û b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @` ÿ? bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Û  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @` ÿ?  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1Nn¬¨Ÿ¨Ì bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1b¢€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ˆ{ûn î a á ‚Â €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á á[[ÑÑ)G*G4½4½>3?3I§I§ST^�^�hiswsw} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±++¥¥ ™ ™"#%‰%‰'ÿ'ÿ)u*u,ë,ë.a/a1×1×3M4M6Ã6Ã8999;¯;¯=%>%@›@›BCE‡E‡GýGýIsJsLçLçN[O[QÏQÏSCTCV·V·X+Y+[Ÿ[Ÿ]^`‡`‡bûbûdoeogãgãiWjWlËlËn?o?q³q³s't'v›v›xy{ƒ{ƒ}÷}÷k€k‚߂߄S…S‡LJlj;Š;Œ¯Œ¯Ž#�#‘—‘—“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1%~!1Ð%~!1þ!1Lþ!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1÷ç!1(÷ç!1•("1¤•("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1ÉQ"1€ÉQ"1g’"1üg’"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1 8#1, 8#1U‘#1˜U‘#1ç$1$ç$1¿ò$1è¿ò$1ùÞ%1ìùÞ%1G�&1ÌG�&1S'1S'1¿e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªb. b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _Ÿy¹þ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªb. b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _Ÿy¹þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªb.  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _Ÿy¹þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 .Nw—ˆ¨ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1’Ò€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1HÈO Ï V Ö ] Ý R €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ã ã\\ÕÕ)N*N4Ç4Ç>@?@I·I·S,T,^Ÿ^Ÿhis…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjãã\\ÕÕ N N Ç Ç@@¹¹22««$$�� � �"#%�%�'ú'ú)s*s,ì,ì.e/e1Þ1Þ3W4W6Ð6Ð8I9I;Â;Â=;>;@´@´B-C-E¦E¦GHJ˜J˜LMO†O†QýQýStTtVëVëXbYb[Ù[Ù]P^P`Ç`Çb'?'I™I™S T ^}^}hïhïrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ��  "÷"÷$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1+~!1Ð+~!1ɾ!1Lɾ!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1ýç!1(ýç!1›("1¤›("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1ÏQ"1€ÏQ"1m’"1üm’"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1ÇÚ"1´ ÇÚ"1'ÿ"10 'ÿ"1i,#1Pi,#1Í:#1,Í:#1’#1˜’#1Õ$1$Õ$1­ò$1è­ò$1çÞ%1ìçÞ%1Ï€&1ÌÏ€&1ŸR'1ŸR'1Ge'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬f2 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _Ÿx¸þ> bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬f2 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _Ÿx¸þ> bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬f2  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _Ÿx¸þ>  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 9Y¢Â…¥ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Y™ €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1d ä k ë ròyùÙ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>2?2I§I§ST^‹^‹hühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbêbêd]e]gÐgÐiCjCl¶l¶n'o'q˜q˜s t vzvzxëxëz\{\}Í}Í>€>‚¯‚¯„ … ‡’‡’‰ŠŒtŒtŽåŽå�V‘V“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lOi!1×}!1Ð×}!1u¾!1Lu¾!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1©ç!1(©ç!1G("1¤G("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1{Q"1€{Q"1’"1ü’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1eÓ"1Ø eÓ"1×Ù"1´ ×Ù"17þ"18 7þ"1y+#1Xy+#1ñ7#14ñ7#1��#1 ��#1É$1,É$1¡ò$1ð¡ò$1ÛÞ%1ôÛÞ%1™€&1Ô™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ü b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @` ÿ? bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ü b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @` ÿ? bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ü  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @` ÿ?  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1ÏïCc–’“· bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1x¸€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1à`Þ^ Ü \ Ú Z Hˆ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚOOÄÄ)9*94®4®>!?!I”I”ST^z^zhíhír`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?M@MBÀBÀD3E3G¦G¦IJLŒLŒNÿNÿPrQrSåSåUXVXXËXËZ>[>]±]±_$`$b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†T‡T‰ƉƋ8Œ8ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lOi!1~!1Ð~!1½¾!1L½¾!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1ñç!1(ñç!1�("1¤�("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1ÃQ"1€ÃQ"1a’"1üa’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1eÓ"1Ø eÓ"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1!8#1,!8#1!’#1˜!’#1ó$1$ó$1Ëò$1èËò$1ß%1ìß%1Y�&1ÌY�&1)S'1)S'1Ñe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸW# b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _ŸŸßÿ? bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸW# b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _ŸŸßÿ? bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸW#  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _ŸŸßÿ?  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 !Aa�ƒ£ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¨è€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1  ² 2 Ä D Ö V Ø €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âZZÒÒ)J*J4Â4Â>:?:I°I°S&T&^œ^œhisˆsˆ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢’’  ‚ ‚"ú"ú$r%r'ê'ê)b*b,Ú,Ú.R/R1Ê1Ê3B4B6º6º8292;ª;ª=">"@š@šBCEŠEŠGHJzJzLðLðNfOfQÜQÜSRTRVÈVÈX>Y>[´[´]*^*` ` bceŒeŒghjxjxlîlîndodqÚqÚsPtPvÆvÆx?IŠIŠSûSû]l^lhÝhÝrNsN} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒÿÿrrååX X"Ë"Ë$>%>'±'±)$*$,—,—. / 1}1}3ð3ð5c6c8Ö8Ö:I;I=¼=¼?-@-BžBžDEG€G€IñIñKbLbNÓNÓPDQDSµSµU&V&X—X—Z[]y]y_ê_êa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lOi!1%~!1Ð%~!1þ!1Lþ!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1÷ç!1(÷ç!1•("1¤•("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1ÉQ"1€ÉQ"1g’"1üg’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1eÓ"1Ø eÓ"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1÷:#1,÷:#1ó’#1˜ó’#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1á€&1Ìá€&1±R'1±R'1Ye'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡[' b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 _ŸŸßÿ? bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡[' b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 _ŸŸßÿ? bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡['  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 _ŸŸßÿ?  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 0P�°„¤ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1zº €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1è h ú z Œžªê €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞTTÊÊ)@*@4¶4¶>,?,I I ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽzzððf f"Ü"Ü$R%R'È'È)>*>,´,´.*/*1 1 346Œ6Œ89;x;x=î=î?d@dBÚBÚDPEPGÆGÆI8?8I¬I¬S T ^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðxxððhhàà X X Ð ÐHHÀÀ88°°%%šš „ „"ù"ù$n%n'ã'ã)X*X,Í,Í.B/B1·1·3,4,6¡6¡89;‹;‹=>@u@uBêBêD_E_GÔGÔIHJHL¼L¼N0O0Q¤Q¤STVŒVŒXY[t[t]è]è_\`\bÐbÐdDeDg¸g¸i,j,l l noqˆqˆsüsüupvpxäxäzX{X}Ì}Ì@€@‚´‚´„(…(‡œ‡œ‰ŠŒ†Œ†ŽûŽû�p‘p“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1~!1�~!1½¾!1 ½¾!1MÅ!1MÅ!1ëç!1€ëç!1‰("1ü‰("1/"1/"1½Q"1p½Q"1[’"1ì[’"1ñ˜"1ôñ˜"1»Ú"1`»Ú"1ÿ"1Ìÿ"1;0#1ì;0#1#1L #1ó$1¬ ó$1Ëò$1p Ëò$1ß%1t ß%1)�&1T )�&1ùR'1˜ùR'1¡e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“K b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Ï€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“K b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Ï€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“K b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Ï €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1k‹  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss ê êaaØØ)O*O4Æ4Æ>;?;I°I°S%T%^š^šhis„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssêêaaØØ O O Æ Æ==´´++¢¢�� ~ ~"õ"õ$l%l'ã'ã)Z*Z,Ñ,Ñ.H/H1¿1¿36466­6­8$9$;›;›=>@‰@‰BCEwEwGîGîIcJcLØLØNMOMQÂQÂS7T7V¬V¬X!Y![–[–] ^ `€`€bõbõdjejgßgßiTjTlÉlÉn>o>q³q³s(t(v�v�xy{‡{‡}ü}üq€q‚æ‚æ„[…[‡ЇЉEŠEŒºŒºŽ/�/‘¤‘¤“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1ï}!1�ï}!1�¾!1�¾!1¿Â!1$¿Â!1Áç!1�Áç!1_("1_("1‘,"1$‘,"1“Q"1�“Q"11’"11’"1c–"1$c–"1Ù"1�Ù"1}ý"1ü}ý"1;0#1 ;0#1#1| #1á$1Ü á$1¹ò$1  ¹ò$1óÞ%1¤ óÞ%1±€&1„ ±€&1�R'1È�R'1)e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14u©êÔ  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14u©êÔ  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14u©êÔ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ì ìaaÖÖ)K*K4À4À>3?3I¦I¦ST^Š^Šhühürnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvììbbØØ N N Ä Ä::°°&&œœ††ûûp p"å"å$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ì=ì?a@aBÖBÖDKEKGÀGÀI3J3L¦L¦NOQŒQŒSÿSÿUrVrXåXåZX[X]Ë]Ë_>`>b°b°d"e"g”g”ijlxlxnênêp\q\sÎsÎu@v@x²x²z${$}–}–€‚z‚z„ì„ì†^‡^‰щыDŒDŽ·Ž·�*‘*“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1~!1�~!1½¾!1 ½¾!1MÅ!1MÅ!1ñç!1€ñç!1�("1ü�("1/"1/"1ÃQ"1pÃQ"1a’"1ìa’"1ñ˜"1ôñ˜"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1;0#1ì;0#1#1L #1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1À&1T À&1“R'1˜“R'1;e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“M b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ŽÎ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“M b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ŽÎ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1_“M b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ŽÎ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1v–  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ê ê``ÕÕ)J*J4¿4¿>2?2I¥I¥ST^‹^‹hþhþrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttêê``ÖÖ L L  Â88®®$$šš††üür r"è"è$]%]'Ò'Ò)G*G,¼,¼.1/11¦1¦346�6�89;z;z=ï=ï?d@dBÙBÙDNENGÃGÃI6J6L©L©NOQ�Q�STVuVuXèXèZ[[[]Î]Î_A`Ab´b´d'e'gšgši j l€l€nónópfqfsÙsÙuLvLx¿x¿z2{2}¥}¥€‚‹‚‹„þ„þ†q‡q‰ä‰ä‹WŒWŽÊŽÊ�=‘=“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1ï}!1�ï}!1�¾!1 �¾!1¿Â!1¿Â!1»ç!1€»ç!1Y("1üY("1‘,"1‘,"1�Q"1p�Q"1+’"1ì+’"1c–"1ôc–"1Ù"1`Ù"1}ý"1Ì}ý"1;0#1ì;0#1#1L #1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1�&1T �&1çR'1˜çR'1�e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨ë b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨ë b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨ë b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ë ëbbÙÙ)P*P4Ç4Ç>>?>IµIµS,T,^ž^žhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttëëbbÙÙ P P Ç Ç>>µµ,,££‘‘  "ö"ö$m%m'ä'ä)[*[,Ò,Ò.I/I1À1À37476®6®8%9%;œ;œ=>@Š@ŠBCExExGïGïIfJfLÝLÝNTOTQËQËSBTBV¹V¹X0Y0[§[§]^`•`•b c eƒeƒgúgúiqjqlèlènZoZqÌqÌs>t>v°v°x"y"{”{”}~€x€x‚ê‚ê„\…\‡·Ή@Š@Œ²Œ²Ž$�$‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1~!1�~!1½¾!1 ½¾!1YÅ!1YÅ!1ñç!1€ñç!1�("1ü�("1+/"1+/"1ÃQ"1pÃQ"1a’"1ìa’"1ý˜"1ôý˜"1Ù"1`Ù"1}ý"1Ì}ý"1;0#1ì;0#1#1L #1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœT  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ÿß€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœT  b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ÿß€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœT  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ÿß €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss í íccÙÙ)O*O4Ã4Ã>7?7I«I«ST^“^“hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssííggáá [ [ Õ ÕOOÉÉCC½½33©© • •" # %�%�'÷'÷)m*m,ã,ã.Y/Y1Ï1Ï3E4E6»6»8191;§;§=>@�@�BCEwEwGëGëI_J_LÓLÓNGOGQ»Q»S/T/V£V£XY[‹[‹]ÿ]ÿ_s`sbçbçd[e[gÏgÏiCjCl·l·n+o+qŸqŸstv‡v‡xûxûzo{o}ã}ãW€W‚˂˄?…?‡±‡±‰$Š$Œ—Œ—Ž � ‘}‘}“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1é}!1�é}!1‡¾!1 ‡¾!1¿Â!1¿Â!1»ç!1€»ç!1Y("1üY("1‘,"1‘,"1�Q"1p�Q"1+’"1ì+’"1c–"1ôc–"1Ù"1`Ù"1}ý"1Ì}ý"1;0#1ì;0#1#1L #1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1½€&1T ½€&1�R'1˜�R'15e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦ç b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦ç b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦ç b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ç ç\\ÑÑ)F*F4»4»>0?0I¥I¥ST^Œ^Œhþhþrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrçç\\ÑÑ F F » »00¥¥��yyîîc c"Ø"Ø$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=ß=ß?T@TBÉBÉD>E>G³G³I(J(L�L�NOQ‡Q‡SüSüUqVqXæXæZ[[[]Ð]Ð_E`Ebºbºd/e/g¤g¤ijlŽlŽnoqrqrsäsäuVvVxÈxÈz:{:}¬}¬€‚�‚�„…‡t‡t‰æ‰æ‹XŒXŽÊŽÊ�<‘<“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1%~!1�%~!1þ!1 þ!1YÅ!1YÅ!1ñç!1€ñç!1�("1ü�("1+/"1+/"1ÃQ"1pÃQ"1a’"1ìa’"1ý˜"1ôý˜"1Ù"1`Ù"1}ý"1Ì}ý"1;0#1ì;0#1#1L #1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœV" b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ÿß€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœV" b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ÿß€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœV" b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ÿß €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ss ë ë__ÓÓ)G*G4¹4¹>+?+I�I�ST^�^�hóhórcsc} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðssëëccÛÛ S S Ë ËCC»»33««““ { {"ï"ï$c%c'×'×)K*K,¿,¿.3/31§1§346�6�89;w;w=é=é?[@[BÍBÍD?E?G±G±I#J#L•L•NOQyQySëSëU]V]XÏXÏZA[A]³]³_%`%b—b—d e g{g{iíiík_l_nÑnÑpCqCsµsµu'v'x™x™z { }}}}ïï�a‚a„Ó„Ó†C‡C‰´‰´‹%Œ%Ž–Ž–�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1ã}!1�ã}!1�¾!1 �¾!1ÅÂ!1ÅÂ!1µç!1€µç!1S("1üS("1—,"1—,"1‡Q"1p‡Q"1%’"1ì%’"1i–"1ôi–"1Ù"1`Ù"1wý"1Ìwý"150#1ì50#1#1L #1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1#�&1T #�&1óR'1˜óR'1›e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥è b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥è b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥è b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt î îhhÞÞ)T*T4Ê4Ê>>?>I²I²S&T&^š^šhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttîîhhââ \ \ Ö ÖPPÊÊDD¾¾88²²,, ¦ ¦" # %–%–' ( *‚*‚,ø,ø.n/n1ä1ä3Z4Z6Ð6Ð8F9F;¼;¼=2>2@¨@¨BCE”E”G H J~J~LòLòNfOfQÚQÚSNTNVÂVÂX6Y6[ª[ª]^`’`’bcezezgîgîibjblÖlÖnJoJq¾q¾s2t2v¦v¦xy{Ž{Ž}~€v€v‚ê‚ê„^…^‡Ò‡Ò‰FŠFŒºŒºŽ.�.‘¢‘¢“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1%~!1�%~!1þ!1 þ!1/Å!1/Å!1÷ç!1€÷ç!1•("1ü•("1/"1/"1ÉQ"1pÉQ"1g’"1ìg’"1Ó˜"1ôÓ˜"1Ù"1`Ù"1wý"1Ìwý"150#1ì50#1#1L #1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1G�&1T G�&1S'1˜S'1¿e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥]) b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1w·€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥]) b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1w·€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥]) b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1w· €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo è èaaÚÚ)S*S4Ì4Ì>C?CIºIºS/T/^¢^¢his†s†} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooèèaaÚÚ S S Ì ÌEE¾¾77°°))¢¢ ” ”" # %†%†'ÿ'ÿ)x*x,ñ,ñ.j/j1ã1ã3\4\6Õ6Õ8N9N;Ç;Ç=@>@@¹@¹B2C2E«E«G$H$J›J›LMO‰O‰QRTwTwVîVîXeYe[Ü[Ü]S^S`Ê`Êb?c?e´e´g)h)jžjžlmo†o†qùqùsltlvßvßxRyR{Å{Å}8~8€«€«‚ƒ…‘…‘‡ˆŠtŠtŒæŒæŽX�X‘ʑʓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1ã}!1�ã}!1�¾!1 �¾!1ÅÂ!1ÅÂ!1¯ç!1€¯ç!1M("1üM("1—,"1—,"1�Q"1p�Q"1’"1ì’"1i–"1ôi–"1Ù"1`Ù"1wý"1Ìwý"150#1ì50#1#1L #1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£ä b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£ä b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£ä b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ê êbbÖÖ)J*J4¾4¾>0?0I¢I¢ST^†^†høhørjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrêêbbÚÚ R R Ê ÊBBºº22ªª""šš Š Š"#%v%v'ê'ê)^*^,Ò,Ò.F/F1º1º3.4.6¢6¢89;Š;Š=þ=þ?r@rBæBæDZEZGÎGÎI@J@L²L²N$O$Q–Q–STVzVzXìXìZ^[^]Ð]Ð_B`Bb´b´d&e&g˜g˜i j l|l|nînîp`q`sÒsÒuDvDx¶x¶z({(}š}š € ‚~‚~„ð„ð†b‡b‰Ô‰Ô‹FŒFޏޏ�*‘*“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1+~!1�+~!1ɾ!1 ɾ!1/Å!1/Å!1÷ç!1€÷ç!1•("1ü•("1/"1/"1ÉQ"1pÉQ"1g’"1ìg’"1Ó˜"1ôÓ˜"1Ù"1`Ù"1wý"1Ìwý"150#1ì50#1#1L #1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥_+ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1w·€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥_+ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1w·€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥_+ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1w· €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10oo æ æ]]ÔÔ)K*K4Â4Â>7?7I¬I¬ST^�^�hisqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðooææ]]ÔÔ K K  Â99°°''žžŒŒ z z"ñ"ñ$h%h'ß'ß)V*V,Í,Í.D/D1»1»32426©6©8 9 ;—;—=>@…@…BüBüDsEsGêGêI_J_LÔLÔNIOIQ¾Q¾S3T3V¨V¨XY[’[’]^`|`|bïbïdbebgÕgÕiHjHl»l»n,o,q�q�stvvxðxðza{a}Ò}ÒC€C‚´‚´„%…%‡•‡•‰ŠŒwŒwŽèŽè�Y‘Y“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ý}!1�Ý}!1{¾!1 {¾!1³Â!1³Â!1¯ç!1€¯ç!1M("1üM("1…,"1…,"1�Q"1p�Q"1’"1ì’"1W–"1ôW–"1£Ú"1`£Ú"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢å b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢å b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢å b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1a�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ã ãXXÍÍ)B*B4µ4µ>(?(I›I›ST^�^�hôhôrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““}}òòg g"Ü"Ü$Q%Q'Æ'Æ);*;,°,°.%/%1š1š346„6„8ù8ù:n;n=á=á?T@TBÇBÇD:E:G­G­I J L“L“NOQyQySìSìU_V_XÒXÒZE[E]¸]¸_+`+bžbždeg„g„i÷i÷kjljnÝnÝpPqPsÃsÃu6v6x©x©z{}�}�€‚u‚u„è„è†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1+~!1�+~!1ɾ!1 ɾ!1_Å!1_Å!1ýç!1€ýç!1›("1ü›("11/"11/"1ÏQ"1pÏQ"1m’"1ìm’"1™"1ô™"1Ù"1`Ù"1qý"1Ìqý"1/0#1ì/0#1û~#1L û~#1í$1¬ í$1Åò$1p Åò$1ÿÞ%1t ÿÞ%1S�&1T S�&1#S'1˜#S'1Ëe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®f2 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¨è€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®f2 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¨è€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®f2 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¨è €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk ã ã[[ÓÓ)K*K4Ã4Ã>9?9I¯I¯S%T%^›^›his‡s‡} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkãã[[ÓÓ K K à Ã;;³³++££““  ƒ ƒ"û"û$s%s'ë'ë)c*c,Û,Û.S/S1Ë1Ë3C4C6»6»8393;«;«=#>#@›@›BCE‹E‹GHJyJyLïLïNeOeQÛQÛSQTQVÇVÇX=Y=[³[³])^)`Ÿ`Ÿbce‹e‹ghjwjwlílíncocqÙqÙsOtOvÅvÅx;y;{±{±}'~'€�€�‚ƒ…‰…‰‡ÿ‡ÿ‰uŠuŒëŒëŽa�a‘בד b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ý}!1�Ý}!1{¾!1 {¾!1³Â!1³Â!1©ç!1€©ç!1G("1üG("1…,"1…,"1{Q"1p{Q"1’"1ì’"1W–"1ôW–"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1·$1¬ ·$1�ò$1p �ò$1ÉÞ%1t ÉÞ%19€&1T 9€&1 R'1˜ R'1±d'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l á b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l á b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1l á b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1nŽ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ã ãVVÉÉ)<*<4­4­>?I�I�ST^q^qhâhârQsQ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““yyìì_ _"Ò"Ò$E%E'¸'¸)+*+,ž,ž./1„1„3÷3÷5j6j8Ý8Ý:P;P=Á=Á?2@2B£B£DEG…G…IöIöKgLgNØNØPIQISºSºU+V+XœXœZ [ ]~]~_ï_ïa`b`dÑdÑfBgBi³i³k$l$n•n•pqswswuèuèwYxYzÊzÊ|;};¬¬�‚„ބކý†ýˆm‰m‹Ý‹Ý�MŽM�½�½’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1+~!1�+~!1ɾ!1 ɾ!1_Å!1_Å!1ýç!1€ýç!1›("1ü›("11/"11/"1ÏQ"1pÏQ"1m’"1ìm’"1™"1ô™"1Ù"1`Ù"1qý"1Ìqý"1/0#1ì/0#1û~#1L û~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®h4 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1©é€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®h4 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1©é€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®h4 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1©é €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10kk á áWWÍÍ)C*C4¹4¹>-?-I¡I¡ST^‰^‰hýhýrqsq} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðkkááWWÍÍ C C ¹ ¹//¥¥‘‘}}óói i"ß"ß$U%U'Ë'Ë)A*A,·,·.-/-1£1£346�6�89;{;{=ñ=ñ?g@gBÝBÝDSESGÉGÉI=J=L±L±N%O%Q™Q™S T V�V�XõXõZi[i]Ý]Ý_Q`QbÅbÅd9e9g­g­i!j!l•l•n o q}q}sñsñuevexÙxÙzM{M}Á}Á5€5‚©‚©„…‡‘‡‘‰ŠŒyŒyŽíŽí�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1×}!1�×}!1u¾!1 u¾!1¿Â!1¿Â!1©ç!1€©ç!1G("1üG("1‘,"1‘,"1{Q"1p{Q"1’"1ì’"1c–"1ôc–"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1Û€&1T Û€&1«R'1˜«R'1Se'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸâ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸâ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸâ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1hˆ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ä äYYÎÎ)C*C4¸4¸>,?,I I ST^ˆ^ˆhühürpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllää\\ÔÔ L L Ä Ä<<´´,,¤¤ŽŽ x x"í"í$b%b'×'×)L*L,Á,Á.6/61«1«3 4 6•6•8 9 ;;=ô=ô?i@iBÞBÞDSESGÈGÈI/?/I¤I¤ST^Ž^Žhisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggÞÞUUÌÌ C C º º11¨¨––  „„ûûr r"é"é$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbébéd^e^gÓgÓiHjHl½l½n2o2q§q§stv‘v‘xy{{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ñ}!1�Ñ}!1o¾!1 o¾!1¿Â!1¿Â!1£ç!1€£ç!1A("1üA("1‘,"1‘,"1uQ"1puQ"1’"1ì’"1c–"1ôc–"1»Ú"1`»Ú"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�Þ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�Þ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1i�Þ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1y™  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj à àUUÊÊ)?*?4´4´>'?'IšIšS T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjààVVÌÌ B B ¸ ¸..¤¤��zzïïd d"Ù"Ù$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¤b¤degˆgˆiúiúklllnÞnÞpPqPsÂsÂu4v4x¦x¦z{}Š}Šüü�n‚n„à„à†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!11~!1�1~!1Ͼ!1 Ͼ!1)Å!1)Å!1è!1€è!1¡("1ü¡("1û."1û."1ÕQ"1pÕQ"1s’"1ìs’"1͘"1ô͘"1Ù"1`Ù"1qý"1Ìqý"1/0#1ì/0#1û~#1L û~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·q= b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1x¸€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·q= b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1x¸€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·q= b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1x¸ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ïï  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh Þ ÞTTÉÉ)>*>4³4³>&?&I™I™S T ^^hòhòrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽzzððf f"Ü"Ü$Q%Q'Æ'Æ);*;,°,°.%/%1š1š346„6„8ù8ù:n;n=ã=ã?X@XBÍBÍDBEBG·G·I*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b¨b¨degŽgŽijltltnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ñ}!1�Ñ}!1o¾!1 o¾!1­Â!1­Â!1£ç!1€£ç!1A("1üA("1,"1,"1uQ"1puQ"1’"1ì’"1Q–"1ôQ–"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1)0#1ì)0#1õ~#1L õ~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœß b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœß b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœß b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1m�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>2?2I©I©S T ^’^’hisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÑLÑNHOHQ¿Q¿S6T6V­V­X$Y$[›[›]^`‰`‰bcewewgîgîiejelÜlÜnNoNqÀqÀs2t2v¤v¤xy{ˆ{ˆ}ú}úl€l‚Þ‚Þ„P…P‡‡‰4Š4Œ¦Œ¦Ž�‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!17~!1�7~!1Õ¾!1 Õ¾!1SÅ!1SÅ!1è!1€è!1¡("1ü¡("1%/"1%/"1ÕQ"1pÕQ"1s’"1ìs’"1÷˜"1ô÷˜"1‘Ú"1`‘Ú"1ñþ"1Ìñþ"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1ç€&1T ç€&1·R'1˜·R'1_e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀxD b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¤ä€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀxD b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¤ä€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀxD b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¤ä €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Xx  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg á áWWÍÍ)C*C4·4·>+?+IŸIŸST^‡^‡hûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggáá[[ÕÕ O O É ÉCC½½77±±''�� ‰ ‰"ÿ"ÿ$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@ƒ@ƒB÷B÷DkEkGßGßISJSLÇLÇN;O;Q¯Q¯S#T#V—V—X Y [[]ó]ó_g`gbÛbÛdOeOgÃgÃi7j7l«l«noq“q“stv{v{xïxïzc{c}×}×K€K‚¿‚¿„3…3‡¥‡¥‰ŠŒ‹Œ‹ŽþŽþ�q‘q“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ë}!1�Ë}!1i¾!1 i¾!1­Â!1­Â!1�ç!1€�ç!1;("1ü;("1,"1,"1oQ"1poQ"1 ’"1ì ’"1Q–"1ôQ–"1ÁÚ"1`ÁÚ"1!ÿ"1Ì!ÿ"1)0#1ì)0#1õ~#1L õ~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÛ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÛ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÛ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1†¦  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Û ÛPPÅÅ):*:4¯4¯>$?$I™I™ST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÛÛPPÅÅ : : ¯ ¯$$™™ƒƒøømmââW W"Ì"Ì$A%A'¶'¶)+*+, , ./1Š1Š3ÿ3ÿ5t6t8é8é:^;^=Ó=Ó?H@HB½B½D2E2G§G§IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!17~!1�7~!1Õ¾!1 Õ¾!1SÅ!1SÅ!1 è!1€ è!1§("1ü§("1%/"1%/"1ÛQ"1pÛQ"1y’"1ìy’"1÷˜"1ô÷˜"1¯Ú"1`¯Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1±$1¬ ±$1‰ò$1p ‰ò$1ÃÞ%1t ÃÞ%1o€&1T o€&1?R'1˜?R'1çd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀzF b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¥å€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀzF b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¥å€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŒÀzF b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¥å €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1q‘  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg ß ßSSÇÇ);*;4­4­>?I‘I‘ST^u^uhçhçrWsW} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggßßWWÏÏ G G ¿ ¿77¯¯''ŸŸ‡‡ûûo o"ã"ã$W%W'Ë'Ë)?*?,³,³.'/'1›1›346ƒ6ƒ8÷8÷:k;k=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b‹b‹dýdýfogoiáiákSlSnÅnÅp7q7s©s©uvx�x�zÿzÿ|q}qãã�U‚U„DŽdž7‡7‰¨‰¨‹ŒŽŠŽŠ�û�û’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Ë}!1�Ë}!1i¾!1 i¾!1¹Â!1¹Â!1—ç!1€—ç!15("1ü5("1‹,"1‹,"1iQ"1piQ"1’"1ì’"1]–"1ô]–"1£Ú"1`£Ú"1ÿ"1Ìÿ"1#0#1ì#0#1ï~#1L ï~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1�&1T �&1ÕR'1˜ÕR'1}e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ü b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ü b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ü b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1jŠ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh â â\\ÒÒ)H*H4¾4¾>2?2I¦I¦ST^�^�hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²,,¦¦  š š"#%Š%Š'(*v*v,ì,ì.b/b1Ø1Ø3N4N6Ä6Ä8:9:;°;°=&>&@œ@œBCEˆEˆGþGþIrJrLæLæNZOZQÎQÎSBTBV¶V¶X*Y*[ž[ž]^`†`†búbúdnengâgâiVjVlÊlÊn@o@q´q´s(t(vœvœxy{„{„}ø}øl€l‚à‚à„T…T‡ȇȉ<Š<Œ°Œ°Ž$�$‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1=~!1�=~!1Û¾!1 Û¾!1Å!1Å!1 è!1€ è!1§("1ü§("1ï."1ï."1ÛQ"1pÛQ"1y’"1ìy’"1Á˜"1ôÁ˜"1—Ú"1`—Ú"1÷þ"1Ì÷þ"1#0#1ì#0#1ï~#1L ï~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%15�&1T 5�&1S'1˜S'1­e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•É�M b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1w·€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•É�M b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1w·€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•É�M b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1w· €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1_  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ü ÜUUÎÎ)G*G4À4À>7?7I®I®S#T#^ ^ his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÜÜUUÎÎ G G À À99²²++¤¤–– ˆ ˆ"#%z%z'ó'ó)l*l,å,å.^/^1×1×3P4P6É6É8B9B;»;»=4>4@­@­B&C&EŸEŸGHJ�J�LMO}O}QôQôSkTkVâVâXYYY[Ð[Ð]G^G`¾`¾b3c3e¨e¨ghj’j’lmo„o„q÷q÷sjtjvÝvÝxPyP{Ã{Ã}6~6€©€©‚ƒ…�…�‡ˆŠrŠrŒäŒäŽV�V‘ȑȓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Å}!1�Å}!1c¾!1 c¾!1³Â!1³Â!1—ç!1€—ç!15("1ü5("1…,"1…,"1iQ"1piQ"1’"1ì’"1W–"1ôW–"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1#0#1ì#0#1ï~#1L ï~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c— Ø b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c— Ø b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c— Ø b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1‘±  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÊÊ)>*>4²4²>$?$I–I–ST^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$j%j'Þ'Þ)R*R,Æ,Æ.:/:1®1®3"4"6–6–8 9 ;~;~=ò=ò?f@fBÚBÚDNENGÂGÂI4J4L¦L¦NOQŠQŠSüSüUnVnXàXàZR[R]Ä]Ä_6`6b¨b¨degŒgŒiþiþkplpnænæpXqXsÊsÊuŒ>ްް�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1=~!1�=~!1Û¾!1 Û¾!1Å!1Å!1è!1€è!1­("1ü­("1ï."1ï."1áQ"1páQ"1’"1ì’"1Á˜"1ôÁ˜"1µÚ"1`µÚ"1ÿ"1Ìÿ"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1À&1T À&1“R'1˜“R'1;e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•ɃO b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1w·€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•ɃO b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1w·€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1•ɃO b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1w· €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1|œ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10cc Ú ÚQQÈÈ)?*?4¶4¶>+?+I I ST^Œ^Œhýhýrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðccÚÚQQÈÈ ? ? ¶ ¶--¤¤’’  €€÷÷n n"å"å$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞISJSLÈLÈN=O=Q²Q²S'T'VœVœXY[†[†]û]û_p`pbãbãdVeVgÉgÉi?I‘I‘ST^ƒ^ƒhöhörgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡þþssèè] ]"Ò"Ò$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=×=×?J@JB½B½D0E0G£G£IJL‰L‰NüNüPoQoSâSâUUVUXÈXÈZ;[;]®]®_!`!b b deg†g†iùiùklllnßnßpRqRsÅsÅu8v8x«x«z{}‘}‘€‚w‚w„ê„ê†[‡[‰͉͋?Œ?ޱޱ�#‘#“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L=i!1=~!1�=~!1Û¾!1 Û¾!1GÅ!1GÅ!1è!1€è!1­("1ü­("1/"1/"1áQ"1páQ"1’"1ì’"1ë˜"1ôë˜"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1ç$1¬ ç$1ñ$1p ñ$1íÀ%1t íÀ%1M�&1T M�&1CQ'1˜CQ'1ëc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒŠV b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¢â€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒŠV b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¢â€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒŠV b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¢â €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ÏË  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ × ×OOÑÑ)I*I4Á4Á>7?7I­I­S#T#^¥^¥his‘s‘} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__××OOÇÇ ? ? · ·//§§——‡‡ÿÿw w"ï"ï$q%q'é'é)a*a,Ù,Ù.Q/Q1É1É3A4A6¹6¹8191;©;©=!>!@™@™BCE‰E‰GHJwJwLíLíNcOcQÙQÙSOTOVÅVÅX;Y;[±[±]'^'`�`�bce‰e‰gÿgÿiujulëlënmomqãqãsYtYvÏvÏxEyE{»{»}1~1€§€§‚ƒ…“…“‡ ˆ ŠŠŒõŒõŽk�k‘á‘á“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L=i!1¿}!1�¿}!1]¾!1 ]¾!1§Â!1§Â!1‘ç!1€‘ç!1/("1ü/("1y,"1y,"1cQ"1pcQ"1’"1ì’"1K–"1ôK–"17Ú"1`7Ú"1—þ"1Ô—þ"1#0#1ô#0#1ï~#1T ï~#1«$1´ «$1Óð$1x Óð$1±À%1| ±À%1-€&1\ -€&1#P'1 #P'1Ëb'1¤€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” Õ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” Õ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1`” Õ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7M'operation.3exchange :z2breasonOperatorInputsvalue7invocations1ÆVµ!  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×NNÁÁ)4*44¥4¥>?I‡I‡ST^u^uhæhærUsU} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡þþqqääW W"Ê"Ê$=%='°'°)#*#,–,–. / 1|1|3ï3ï5b6b8Õ8Õ:H;H=¹=¹?*@*B›B›D E G}G}IîIîK_L_NÐNÐPAQAS²S²U#V#X”X”Z[]v]v_ç_çadbddÕdÕfFgFi·i·k(l(n™n™p q s{s{uìuìw]x]zÎzÎ|?}?°°�!‚!„’„’†‡‰q‰q‹á‹á�QŽQ�Á�Á’ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L=i!1C~!1�C~!1á¾!1 á¾!1GÅ!1GÅ!1è!1€è!1­("1ü­("1/"1/"1áQ"1páQ"1’"1ì’"1ë˜"1ôë˜"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1Ï$1¬ Ï$1÷ð$1p ÷ð$1ÕÀ%1t ÕÀ%1Õ€&1T Õ€&1ËP'1˜ËP'1sc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒŒX b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1£ã€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒŒX b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1£ã€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1žÒŒX b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1£ã €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ÏË  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10__ Õ ÕKKÉÉ)?*?4µ4µ>)?)I�I�ST^‘^‘hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ð__ÕÕKKÁÁ 7 7 ­ ­##™™……ûûqqçç] ]"Ó"Ó$Q%Q'Ç'Ç)=*=,³,³.)/)1Ÿ1Ÿ346‹6‹89;w;w=í=í?c@cBÙBÙDOEOGÅGÅI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noq…q…sùsùumvmxáxázU{U}É}É=€=‚±‚±„%…%‡™‡™‰ Š Œ�Œ�ŽõŽõ�i‘i“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„1i!1q}!1q}!1¾!1|¾!1sÃ!1„sÃ!1©Æ!1ð©Æ!13Ð!1|3Ð!1iÓ!1HiÓ!1Cç!1ÔCç!1á'"1Pá'"1E-"1XE-"1{0"1Ä{0"1:"1P:"1;="1;="1Q"1¨Q"1³‘"1$ ³‘"1—"1, —"1Mš"1˜ Mš"1×£"1$ ×£"1 §"1ð  §"1OÔ"1| OÔ"1éÙ"1XéÙ"1Iþ"1ÜIþ"1‹+#1ü‹+#1©:#1Ø©:#1eo#1deo#1��#1ˆ��#1‹š#1´‹š#1Ï$1@Ï$1§ò$1§ò$1áÞ%1áÞ%1[ì%1è[ì%1ç€&1tç€&1·R'1¸·R'1_e'1¼€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PÉ• b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ñ bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PÉ• b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ñ  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1PÉ•  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ñ bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ‰…Ž-á bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1@b¢C¥åG‡  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d€bâÉICÃ’$¤[Û…æfÇG¨(‰ I‰këMÍ/ ¯  ‘ ó s Õ U · 7 ™ zº[½ý_Ÿ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Û Û  ÑÑ)G*G4½4½>1?1I¥I¥S%T%^™^™h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÛÛUUÏÏ I I à Ã==··11««ÝÝSSÉÉ? ?"µ"µ$y%y'ï'ï)e*e,Û,Û.Q/Q1Ç1Ç3=4=6³6³8)9);Ÿ;Ÿ=>@‰@‰BýBýDqEqGåGåIYJYLÍLÍNAOAQµQµS)T)V�V�XY[…[…]ù]ù_m`mbíbídaeagÕgÕiIjIl½l½n1o1q¥q¥stv�v�xy{u{u}é}é]€]‚тфE…E‡¹‡¹‰,Š,ŒŸŒŸŽ�‘…‘…“# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1U~!1èU~!1ó¾!1dó¾!1§Å!1l§Å!11Ï!1è1Ï!1gÒ!1´gÒ!1'è!1@'è!1Å("1¼Å("1y/"1Äy/"19"1@9"19<"1 9<"1ùQ"1˜ùQ"1—’"1—’"1K™"1 K™"1Õ¢"1˜ Õ¢"1 ¦"1d  ¦"1MÓ"1ð MÓ"1Ú"1Ì Ú"1sþ"1H sþ"1µ+#1hµ+#1Ç:#1DÇ:#1ƒo#1Ѓo#1�#1ô�#1›#1 ›#1·$1¬·$1�ò$1p�ò$1ÉÞ%1tÉÞ%1Cì%1TCì%1�€&1à�€&1]R'1$]R'1e'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ºî¨t b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡]�B bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ºî¨t b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡]�B bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ºî¨t  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡]�B  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ^~Ññì  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1@g§N·÷`   bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d€gçÇGNÎŽ5µVÖœƒêjÑQ¸8¡!Š s ó \ Ü E Å . ® c £  — €ç'ŽÎ7wà  bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×  ÃÃ)8*84­4­> ? I“I“ST^…^…høhørksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ºº//¤¤ "Ž"Ž$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?L@LB¿B¿D2E2G¥G¥IJL‹L‹NþNþPqQqSäSäUWVWXÊXÊZ=[=]°]°_#`#b¢b¢degˆgˆiûiûknlnnánápTqTsÇsÇu:v:x­x­z { }“}“€‚y‚y„ì„ì†_‡_‰щыCŒC޵޵�'‘'“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„1i!1q}!1q}!1¾!1|¾!1sÃ!1„sÃ!1©Æ!1ð©Æ!13Ð!1|3Ð!1iÓ!1HiÓ!1=ç!1Ô=ç!1Û'"1PÛ'"1E-"1XE-"1{0"1Ä{0"1:"1P:"1;="1;="1Q"1¨Q"1­‘"1$ ­‘"1—"1, —"1Mš"1˜ Mš"1×£"1$ ×£"1 §"1ð  §"1KÕ"1| KÕ"1[Ú"1T[Ú"1»þ"1лþ"1ù,#1ðù,#1';#1È';#1ão#1Tão#1ñ‘#1xñ‘#1ïœ#1¤ïœ#1·$10·$1�ò$1ô�ò$1ÉÞ%1øÉÞ%1Cì%1ØCì%1o€&1do€&1?R'1¨?R'1çd'1¬€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OÆ’ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Ñ bšWuoperation.0ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue4bytes2304patternContiguousinvocations1 bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OÆ’ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Ñ  bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1  bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1OÆ’  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Ñ bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ok-Mz bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Á‰ÍB‚AÀ  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d�‚ƒ„%¥în…·7†€‡Òˆ‡ †  …  „  ƒ  ‚ �€�ÁB‚AÀ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ù ÙÉÉ)=*=4±4±>#?#I•I•ST^…^…h÷h÷risi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A ¹ ¹11©©!!™™ÈÈ<<°°$ $"˜"˜$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=á=á?S@SBÅBÅD7E7G©G©IJL�L�NÿNÿPqQqSãSãUUVUXÇXÇZ9[9]«]«_`b›b›d e ggiñiñkclcnÕnÕpGqGs¹s¹u+v+x�x�z{}�}�óó�e‚e„ׄ׆I‡I‰º‰º‹+Œ+ŽœŽœ� ‘ “# b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 $ bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx1i!1[~!1è[~!1ù¾!1dù¾!1§Å!1l§Å!11Ï!1è1Ï!1gÒ!1´gÒ!1-è!1@-è!1Ë("1¼Ë("1y/"1Äy/"19"1@9"19<"1 9<"1ÿQ"1˜ÿQ"1�’"1�’"1K™"1 K™"1Õ¢"1˜ Õ¢"1 ¦"1d  ¦"1MÓ"1ð MÓ"1OÚ"1Ì OÚ"1¯þ"1H ¯þ"1ñ+#1hñ+#1K;#1DK;#1p#1Ðp#1¥’#1ô¥’#1£�#1 £�#1¥$1¬¥$1}ò$1p}ò$1·Þ%1t·Þ%11ì%1T1ì%1'€&1à'€&1÷Q'1$÷Q'1Ÿd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¼ð¬x b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡^žB bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¼ð¬x b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡^žB bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¼ð¬x  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡^žB  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ~žÑñ(H bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1ŠÊ‡ÇT”\ä$  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1dŠ ”ž�¨(äd²2«+¼<ròÆFÐPØX à ` è h ð p ø x ê * €ˆ�Zä$¬ìt´ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×  ÁÁ)4*44§4§>?I‰I‰ST^w^whèhèrYsY} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡¹¹,,ŸŸ "…"…$=%='°'°)#*#,–,–. / 1|1|3ï3ï5b6b8Õ8Õ:H;H=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éafbfd×d×fHgHi¹i¹k*l*n›n›p q s}s}uîuîw_x_zÐzÐ|A}A²²�#‚#„”„”†‡‰u‰u‹å‹å�UŽU�Å�Å’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx7i!1¿}!1è¿}!1]¾!1d]¾!1§Å!1l§Å!11Ï!1è1Ï!1gÒ!1´gÒ!1‘ç!1@‘ç!1/("1¼/("1y/"1Äy/"19"1@9"19<"1 9<"1cQ"1˜cQ"1’"1’"1K™"1 K™"1Õ¢"1˜ Õ¢"1 ¦"1d  ¦"1MÓ"1ð MÓ"1ãÙ"1Ì ãÙ"1Cþ"1P Cþ"1…+#1p…+#1¯:#1L¯:#1ko#1Øko#1‡�#1ü‡�#1…š#1(…š#1Õ$1´Õ$1­ò$1x­ò$1çÞ%1|çÞ%1aì%1\aì%15�&1è5�&1S'1,S'1­e'10€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ú b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @b¢C bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ú b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @b¢C bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Ú  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @b¢C  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ†‚Ž$Ø bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1@b¢C¤äE…  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d€bâÈHCÃ�$¤XØ…æfÇG¨(‰ @€jêKË, ¬  � î n Ï O ° 0 ‘ r²S´ôU• bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ú ÚSSÌÌ)E*E4¾4¾>7?7I®I®SßSß] ^ his†s†} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÚÚSSÌÌ E E ¾ ¾77°°))¢¢””  † †"ÿ"ÿ$x%x'ñ'ñ)j*j,ã,ã.\/\1Õ1Õ3N4N6Ç6Ç8@9@;¹;¹=2>2@«@«B$C$E�E�GHJ�J�LMO}O}QôQôSkTkVâVâXYYY[Ð[Ð]G^G`¾`¾bïbïddedgÙgÙiNjNlÃlÃn„o„q÷q÷sjtjvÝvÝxPyP{Ã{Ã}6~6€©€©‚ƒ…�…�‡ˆŠtŠtŒæŒæŽX�X‘ʑʓ  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx7i!1O~!1èO~!1í¾!1dí¾!1§Å!1l§Å!11Ï!1è1Ï!1gÒ!1´gÒ!1!è!1@!è!1¿("1¼¿("1y/"1Äy/"19"1@9"19<"1 9<"1óQ"1˜óQ"1‘’"1‘’"1K™"1 K™"1Õ¢"1˜ Õ¢"1 ¦"1d  ¦"1MÓ"1ð MÓ"1Ú"1Ì Ú"1gþ"1H gþ"1©+#1h©+#1¯:#1D¯:#1ko#1Ðko#1‡�#1ô‡�#1…š#1 …š#1½$1¬½$1•ò$1p•ò$1ÏÞ%1tÏÞ%1Iì%1TIì%1É€&1àÉ€&1™R'1$™R'1Ae'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¯ã�i b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡³óB bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¯ã�i b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡³óB bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¯ã�i  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡³óB  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 UuÑñÚú bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1@b¢C¤äE…  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1d€bâÒRCä$$¤vö…æfÇG¨(‰ jêKË, ¬  � î n Ï O ° 0  P ‘ r²S´ôU• bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ü ÜQQÆÆ);*;4°4°>&?&IšIšSÌSÌ]„^„høhørnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÜÜTTÌÌ D D ¼ ¼44¬¬$$œœ††ûûp p"å"å$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ì=ì?a@aBÖBÖDKEKGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb|b|dðdðfdgdiØiØkLlLnoqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡އމŠŒxŒxŽíŽí�b‘b“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx7i!1¿}!1è¿}!1]¾!1d]¾!1§Å!1l§Å!11Ï!1è1Ï!1gÒ!1´gÒ!1‘ç!1@‘ç!1/("1¼/("1y/"1Äy/"19"1@9"19<"1 9<"1cQ"1˜cQ"1’"1’"1K™"1 K™"1Õ¢"1˜ Õ¢"1 ¦"1d  ¦"1IÔ"1ð IÔ"1…Ú"1È …Ú"1åþ"1D åþ"1#-#1d#-#1¯:#1<¯:#1ko#1Èko#1‡�#1ì‡�#1…š#1…š#1Ã$1¤Ã$1›ò$1h›ò$1ÕÞ%1lÕÞ%1Oì%1LOì%1½€&1ؽ€&1�R'1�R'15e'1 €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a• Ø b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @b¢C bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a• Ø b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @b¢C bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a• Ø  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @b¢C  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 fbh°Rr bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1a¡ˆÌB£ãD„  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1daáBÂ#£„  åeèhÆF°0§'xøˆÈiéJÊ+ «  Œ í m Î N ¯ / � q±R³óT” bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>+?+I I SÎSÎ]Œ^Œhýhýroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜISJSLÈLÈN=O=Q²Q²S'T'VœVœXY[†[†]û]û_p`pbžbždeg„g„i÷i÷kjljn(o(q™q™s t v{v{xìxìz]{]}Î}Î?€?‚°‚°„!…!‡“‡“‰ŠŒuŒu޿޿�W‘W“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx7i!1U~!1èU~!1ó¾!1dó¾!1§Å!1l§Å!11Ï!1è1Ï!1gÒ!1´gÒ!1!è!1@!è!1¿("1¼¿("1y/"1Äy/"19"1@9"19<"1 9<"1óQ"1˜óQ"1‘’"1‘’"1K™"1 K™"1Õ¢"1˜ Õ¢"1 ¦"1d  ¦"1MÓ"1ð MÓ"1=Ú"1Ì =Ú"1�þ"1H �þ"1ß+#1hß+#1¯:#1D¯:#1ko#1Ðko#1‡�#1ô‡�#1…š#1 …š#1±$1¬±$1‰ò$1p‰ò$1ÃÞ%1tÃÞ%1=ì%1T=ì%1‡€&1à‡€&1WR'1$WR'1ÿd'1(€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1±å¡m b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡²òB bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1±å¡m b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡²òB bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1±å¡m  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡²òB  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 u•Ññ2 bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7MWoperation.4exchange :z2breasonOperatorInputsvalue8invocations1a¡’ÒB£ãD„  bâ-g operation.4ipu_stack_rearrange_row_major_to_amp_f16_o1_r64_p64_c72_p80-:-z1211Z5ò¡z¥*¥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€� bJ7M—operation.5exchange :z2breasonOperatorInputsvalue9invocations1daáBÂ#£„HÈåešÆFìl§'¾>ˆiéJÊ+ «  Œ í m Î N ¯ / � ¢ â q±R³óT” bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØMMÂÂ)7*74¬4¬>!?!I”I”SÅSÅ]|^|hîhîrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆýýrrçç\ \"Ñ"Ñ$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?M@MBÂBÂD7E7G¬G¬I!J!L”L”NOQzQzSíSíU`V`XÓXÓZF[F]¹]¹_,`,b]b]dÏdÏfAgAi³i³k%l%nÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†R‡R‰ʼnŋ8Œ8ޫޫ�‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1Å}!1ÐÅ}!1c¾!1Lc¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1—ç!1(—ç!15("1¤5("1/"1¬/"1 9"1( 9"1?<"1ô?<"1iQ"1€iQ"1’"1ü’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1SÓ"1Ø SÓ"1ÑÙ"1´ ÑÙ"11þ"18 1þ"1s+#1Xs+#1i2#14i2#1y…#1 y…#1Õ$1,Õ$1­ò$1ð­ò$1çÞ%1ôçÞ%1;�&1Ô;�&1 S'1 S'1³e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–Û b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Ua¡B bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–Û b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Ua¡B bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b–Û  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Ua¡B  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ{w|  À bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ÉI’[ÛHˆ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ù ÙQQÉÉ)A*A4¹4¹>1?1I§I§ST^“^“h i ss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A ¹ ¹11©©!!™™‰‰ y y"ñ"ñ$i%i'á'á)Y*Y,Ñ,Ñ.I/I1Á1Á39496±6±8)9);¡;¡=>@‘@‘B C E�E�GùGùIqJqLçLçN]O]QÓQÓSITIV¿V¿X5Y5[«[«]!^!`—`—b c eƒeƒgùgùiojolålån[o[qÑqÑsGtGv½v½x3y3{©{©}~€•€•‚ ƒ …�…�‡÷‡÷‰mŠmŒãŒãŽY�Y‘ϑϓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1I~!1ÐI~!1ç¾!1Lç¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1è!1(è!1¹("1¤¹("1…/"1¬…/"19"1(9"1E<"1ôE<"1íQ"1€íQ"1‹’"1ü‹’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1õÙ"1´ õÙ"1Uþ"10 Uþ"1—+#1P—+#1i2#1,i2#1óŒ#1˜óŒ#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1ó€&1Ìó€&1ÃR'1ÃR'1ke'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¤Ø’^ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡uµF bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¤Ø’^ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡uµF bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¤Ø’^  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡uµF  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 LlÑñÈè bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€Ý]º:—À  €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Û ÛRRÉÉ)@*@4·4·>0?0I§I§ST^�^�histst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÛÛRRÉÉ @ @ · ·..¥¥““  ��øøo o"æ"æ$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIXJXLÏLÏNFOFQ½Q½S4T4V«V«X"Y"[™[™]^`‡`‡bþbþdueugìgìicjclÚlÚnLoLq¾q¾s0t0v¢v¢xy{†{†}ø}øj€j‚܂܄N…N‡À‡À‰2Š2Œ¤Œ¤Ž�‘ˆ‘ˆ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1Å}!1ÐÅ}!1c¾!1Lc¾!1­Å!1T­Å!17Ï!1Ð7Ï!1mÒ!1œmÒ!1‘ç!1(‘ç!1/("1¤/("1/"1¬/"1 9"1( 9"1?<"1ô?<"1cQ"1€cQ"1’"1ü’"1Q™"1 Q™"1Û¢"1€ Û¢"1¦"1L ¦"1OÔ"1Ø OÔ"1©Ú"1° ©Ú"1 ÿ"1,  ÿ"1G-#1LG-#1õ3#1$õ3#1†#1�†#1½$1½$1•ò$1à•ò$1ÏÞ%1äÏÞ%1À&1ÄÀ&1“R'1“R'1;e'1 €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– Ù b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 Va¡B bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– Ù b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 Va¡B bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– Ù  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 Va¡B  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 [WWŸu• bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1‰Í€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1$¤ím¶6ÿÑ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×MMÃÃ)9*94¯4¯>%?%I™I™S T ^�^�hõhõrisi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$K%K'Á'Á)7*7,­,­.#/#1™1™346…6…8û8û:q;q=ç=ç?]@]BÓBÓDIEIG¿G¿I5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`Ib½b½d1e1g¥g¥ijl�l�noququséséu]v]xÑxÑzE{E}¹}¹-€-‚¡‚¡„…‡‰‡‰‰ý‰ý‹qŒqŽåŽå�Y‘Y“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1O~!1ÐO~!1í¾!1Lí¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1è!1(è!1¹("1¤¹("1…/"1¬…/"19"1(9"1E<"1ôE<"1íQ"1€íQ"1‹’"1ü‹’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1+Ú"1´ +Ú"1‹þ"10 ‹þ"1Í+#1PÍ+#15#1,5#1��#1˜��#1·$1$·$1�ò$1è�ò$1ÉÞ%1ìÉÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¦Ú–b b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡uµG bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¦Ú–b b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡uµG bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1¦Ú–b  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡uµG  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 lŒÑñü bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1�Ý€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1tôQÑ.® ‹]� €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb × ×LLÁÁ)6*64«4«>"?"I—I—S T ^~^~hðhðrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbb××LLÁÁ 6 6 « «  ••  ôôiiÞÞS S"È"È$=%='²'²)'*',œ,œ./1†1†3û3û5p6p8å8å:Z;Z=Ï=Ï?D@DB¹B¹D.E.G£G£IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1Ë}!1ÐË}!1i¾!1Li¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1�ç!1(�ç!1;("1¤;("1…/"1¬…/"19"1(9"1E<"1ôE<"1oQ"1€oQ"1 ’"1ü ’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1YÓ"1Ø YÓ"1ÅÙ"1´ ÅÙ"1%þ"18 %þ"1g+#1Xg+#1o2#14o2#1Ù…#1 Ù…#1Ï$1,Ï$1§ò$1ð§ò$1áÞ%1ôáÞ%1�&1Ô�&1ÕR'1ÕR'1}e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ü b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @c£D bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ü b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @c£D bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ü  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @c£D  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññrnnþþ² bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1B€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1‚ÖVª*~þ¢â €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>+?+I I ST^Š^Šhÿhÿrtst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜISJSLÈLÈN=O=Q²Q²S'T'VœVœXY[†[†]û]û_p`pbåbådZeZgÏgÏiDjDl¹l¹n.o.q£q£stv�v�xy{w{w}ì}ìa€a‚Ö‚Ö„K…K‡À‡À‰5Š5ŒªŒªŽ�‘”‘”“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1C~!1ÐC~!1á¾!1Lá¾!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1è!1(è!1­("1¤­("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1áQ"1€áQ"1’"1ü’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1eÓ"1Ø eÓ"1éÙ"1´ éÙ"1Iþ"10 Iþ"1‹+#1P‹+#1i2#1,i2#1§�#1˜§�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1™Í‡S b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡¨èQ bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1™Í‡S b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡¨èQ bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1™Í‡S  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡¨èQ  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ÑñCc¶Ö bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€èhÐP¸8p° €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh â â\\ÒÒ)H*H4¾4¾>4?4I¨I¨ST^�^�hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhââ\\ÖÖ P P Ê ÊDD¾¾88²²,,¦¦  š š"#%Š%Š'(*v*v,ì,ì.b/b1Ø1Ø3N4N6Ä6Ä8:9:;°;°=&>&@œ@œBCEˆEˆGþGþItJtLèLèN\O\QÐQÐSDTDV¸V¸X,Y,[ [ ]^`ˆ`ˆbübüdpepgägäiXjXlÌlÌn@o@q´q´s(t(vœvœxy{„{„}ø}øl€l‚à‚à„T…T‡ȇȉ<Š<Œ°Œ°Ž$�$‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l=i!1Ë}!1ÐË}!1i¾!1Li¾!1³Å!1T³Å!1=Ï!1Ð=Ï!1sÒ!1œsÒ!1—ç!1(—ç!15("1¤5("1…/"1¬…/"19"1(9"1E<"1ôE<"1iQ"1€iQ"1’"1ü’"1W™"1 W™"1á¢"1€ á¢"1¦"1L ¦"1UÔ"1Ø UÔ"1µÚ"1° µÚ"1ÿ"1, ÿ"1S-#1LS-#14#1$4#1o†#1�o†#1½$1½$1•ò$1à•ò$1ÏÞ%1äÏÞ%1Ÿ€&1ÄŸ€&1oR'1oR'1e'1 €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ú b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @b¢C bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ú b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @b¢C bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1c—Ú  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @b¢C  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 PLL”zš bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1–Ú€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1RÒ&¦úzÎN6v €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ø ØNNÃÃ)8*84­4­>"?"I•I•ST^{^{hîhîrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbØØNNÄÄ : : ° °&&œœˆˆþþttêê` `"Ö"Ö$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?R@RBÇBÇD*>4²4²>&?&I˜I˜S T ^|^|hîhîr`s`} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$j%j'Þ'Þ)R*R,Æ,Æ.:/:1®1®3"4"6–6–8 9 ;~;~=ò=ò?f@fBÚBÚDNENGÂGÂI6J6L¨L¨NOQŒQŒSþSþUpVpXâXâZT[T]Æ]Æ_8`8bªbªdegŽgŽijlrlrnänäpVqVsÈsÈu:v:x¬x¬z{}�}�€‚t‚t„æ„æ†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1Ñ}!1ÐÑ}!1o¾!1Lo¾!1ËÅ!1TËÅ!1UÏ!1ÐUÏ!1‹Ò!1œ‹Ò!1�ç!1(�ç!1;("1¤;("1�/"1¬�/"1'9"1('9"1]<"1ô]<"1oQ"1€oQ"1 ’"1ü ’"1o™"1 o™"1ù¢"1€ ù¢"1/¦"1L /¦"1qÓ"1Ø qÓ"1µÚ"1´ µÚ"1ÿ"18 ÿ"1W,#1XW,#1o2#14o2#1ý…#1 ý…#1Ï$1,Ï$1§ò$1ð§ò$1áÞ%1ôáÞ%1ç€&1Ôç€&1·R'1·R'1_e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Ý b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Iy¹Z bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Ý b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Iy¹Z bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Ý  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Iy¹Z  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1eaañzšñ¥ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ØX°0ˆÀ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee ß ßUUËË)A*A4·4·>+?+IŸIŸST^‡^‡hûhûroso} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeßßYYÓÓ M M Ç ÇAA»»55¯¯%%›› ‡ ‡"ý"ý$s%s'é'é)_*_,Õ,Õ.K/K1Á1Á37476­6­8#9#;™;™=>@ƒ@ƒB÷B÷DkEkGßGßISJSLÇLÇN;O;Q¯Q¯S#T#V—V—X Y [[]ó]ó_g`gbÛbÛdOeOgÃgÃi7j7l«l«noq“q“stv{v{xïxïzc{c}×}×K€K‚¿‚¿„3…3‡§‡§‰ŠŒ�Œ�Ž�‘s‘s“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1=~!1Ð=~!1Û¾!1LÛ¾!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1 è!1( è!1§("1¤§("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1ÛQ"1€ÛQ"1y’"1üy’"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1¯Ú"1´ ¯Ú"1ÿ"10 ÿ"1Q,#1PQ,#1o2#1,o2#1[Ž#1˜[Ž#1½$1$½$1•ò$1è•ò$1ÏÞ%1ìÏÞ%1“€&1Ì“€&1cR'1cR'1 e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂ|H b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡uµE bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂ|H b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡uµE bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂ|H  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡uµE  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 :Z¤Äq‘ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ósæfÙY ` €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛPPÅÅ):*:4¯4¯>"?"I•I•ST^{^{hîhîrasa} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹uuêê_ _"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?N@NBÁBÁD4E4G§G§IJL�L�NOQsQsSæSæUYVYXÌXÌZ?[?]²]²_%`%b˜b˜d e g~g~iñiñkdldn×n×pJqJs½s½u0v0x£x£z{}‰}‰üü�o‚o„â„â†U‡U‰ljNj9Œ9ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1Ë}!1ÐË}!1i¾!1Li¾!1ËÅ!1TËÅ!1UÏ!1ÐUÏ!1‹Ò!1œ‹Ò!1�ç!1(�ç!1;("1¤;("1�/"1¬�/"1'9"1('9"1]<"1ô]<"1oQ"1€oQ"1 ’"1ü ’"1o™"1 o™"1ù¢"1€ ù¢"1/¦"1L /¦"1mÔ"1Ø mÔ"1�Ú"1° �Ú"1ýþ"1, ýþ"1;-#1L;-#1 4#1$ 4#1“†#1�“†#1·$1·$1�ò$1à�ò$1ÉÞ%1äÉÞ%1o€&1Äo€&1?R'1?R'1çd'1 €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Û b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1  Jx¸Y bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Û b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1  Jx¸Y bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Û  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1  Jx¸Y  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 EAA‰cƒ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1˜Ü€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1`à8¸�èhX˜ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝQQÅÅ)9*94­4­>?I‘I‘ST^u^uhçhçrYsY} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��……ùùm m"á"á$U%U'É'É)=*=,±,±.%/%1™1™3 4 6�6�8õ8õ:i;i=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b‹b‹dýdýfogoiáiákSlSnÅnÅp7q7s©s©uvx�x�zÿzÿ|q}qãã�U‚U„DŽdž9‡9‰ª‰ª‹ŒŽŒŽŒ�ý�ý’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lCi!1=~!1Ð=~!1Û¾!1LÛ¾!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1è!1(è!1­("1¤­("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1áQ"1€áQ"1’"1ü’"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1—Ú"1´ —Ú"1÷þ"10 ÷þ"19,#1P9,#1!5#1,!5#1�#1˜�#1«$1$«$1ƒò$1èƒò$1½Þ%1ì½Þ%1-€&1Ì-€&1ýQ'1ýQ'1¥d'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ä€L b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡uµF bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ä€L b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡uµF bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1�Ä€L  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡uµF  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ZzÐðXx bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1³ó€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ÌL¿?²2¥%Ó €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛNNÁÁ)4*44§4§>?I‰I‰SúSú]k^khÜhÜrMsM} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹þþqqääW W"Ê"Ê$=%='°'°)#*#,–,–. / 1|1|3ï3ï5b6b8Õ8Õ:H;H=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éaZbZdËdËf;?;I²I²S'T'^š^šh i s€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÞÞWWÐÐ I I  Â;;´´--¦¦˜˜ Š Š"#%|%|'õ'õ)n*n,ç,ç.`/`1Ù1Ù3R4R6Ë6Ë8D9D;½;½=6>6@¯@¯B(C(E¡E¡GHJ“J“L M O�O�QøQøSoToVæVæX]Y][Ô[Ô]K^K`Â`Âb7c7e¬e¬g!h!j–j–l m o~o~qñqñsdtdv×v×xJyJ{½{½}0~0€£€£‚ƒ…‰…‰‡ü‡ü‰nŠnŒàŒàŽR�R‘đē b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!11~!1Ð1~!1Ͼ!1LϾ!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1è!1(è!1¡("1¤¡("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1ÕQ"1€ÕQ"1s’"1üs’"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1u2#1,u2#1�#1˜�#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1Û€&1ÌÛ€&1«R'1«R'1Se'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·q= b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡¡áC bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·q= b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡¡áC bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ƒ·q=  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡¡áC  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 1Q’²Ž® bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€þ~ü|úzÐ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh à àUUÊÊ)?*?4´4´>*?*IžIžST^†^†húhúrpsp} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhààXXÐÐ H H À À88°°((  ŠŠÿÿt t"é"é$^%^'Ó'Ó)H*H,½,½.2/21§1§346‘6‘89;{;{=ð=ð?e@eBÚBÚDOEOGÄGÄI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡�‡�‰ŠŒzŒzŽïŽï�d‘d“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1Ñ}!1ÐÑ}!1o¾!1Lo¾!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1£ç!1(£ç!1A("1¤A("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1uQ"1€uQ"1’"1ü’"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1[Ô"1Ø [Ô"1ÁÚ"1° ÁÚ"1!ÿ"1, !ÿ"1_-#1L_-#14#1$4#1ù†#1�ù†#1Ï$1Ï$1§ò$1à§ò$1áÞ%1äáÞ%1É€&1ÄÉ€&1™R'1™R'1Ae'1 €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ü b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @b¢C bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ü b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @b¢C bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ü  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @b¢C  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 :66~€  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1£ç€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1Œ oïRÒ5µ»û €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>/?/I¤I¤ST^ˆ^ˆhùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbçbçdZeZgÍgÍi@j@l³l³n$o$q•q•stvwvwxèxèzY{Y}Ê}Ê;€;‚¬‚¬„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!17~!1Ð7~!1Õ¾!1LÕ¾!1¹Å!1T¹Å!1CÏ!1ÐCÏ!1yÒ!1œyÒ!1 è!1( è!1§("1¤§("1‹/"1¬‹/"19"1(9"1K<"1ôK<"1ÛQ"1€ÛQ"1y’"1üy’"1]™"1 ]™"1ç¢"1€ ç¢"1¦"1L ¦"1_Ó"1Ø _Ó"1»Ú"1´ »Ú"1ÿ"10 ÿ"1],#1P],#135#1,35#1Ï�#1˜Ï�#1Ã$1$Ã$1›ò$1è›ò$1ÕÞ%1ìÕÞ%1™€&1Ì™€&1iR'1iR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹uA b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡¡áC bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹uA b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡¡áC bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹uA  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡¡áC  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 QqºÚ{› bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¾þ€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1øxövôtòrŽÎ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜQQÆÆ);*;4°4°>%?%I˜I˜S T ^|^|hîhîrbsb} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒvvëë` `"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1×}!1Ð×}!1u¾!1Lu¾!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1©ç!1(©ç!1G("1¤G("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1{Q"1€{Q"1’"1ü’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1eÓ"1Ø eÓ"1ÇÚ"1´ ÇÚ"1'ÿ"18 'ÿ"1i,#1Xi,#1u2#14u2#1±†#1 ±†#1ç$1,ç$1¿ò$1ð¿ò$1ùÞ%1ôùÞ%1M�&1ÔM�&1S'1S'1Åe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšß b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @b¢C bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšß b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @b¢C bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšß  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @b¢C  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1OKKÛ„¤ëŸ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€înÜ\ÊJp° €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝUUÍÍ)E*E4½4½>5?5I«I«S!T!^—^—h i sƒsƒ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%���� } }"õ"õ$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=>@•@•B C E…E…GýGýIuJuLëLëNaOaQ×Q×SMTMVÃVÃX9Y9[¯[¯]%^%`›`›bce‡e‡gýgýisjslélén_o_qÕqÕsKtKvÁvÁx7y7{­{­}#~#€™€™‚ƒ…………‡û‡û‰qŠqŒçŒçŽ]�]‘Ó‘Ó“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1+~!1Ð+~!1ɾ!1Lɾ!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1ýç!1(ýç!1›("1¤›("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1ÏQ"1€ÏQ"1m’"1üm’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1eÓ"1Ø eÓ"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1u2#1,u2#1Ã�#1˜Ã�#1Û$1$Û$1³ò$1è³ò$1íÞ%1ìíÞ%1 �&1Ì �&1ÛR'1ÛR'1ƒe'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬f2 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡w·B bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬f2 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡w·B bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1x¬f2  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡w·B  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 (H€ ‡§ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ ‰’›€À €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>4?4I«I«S"T"^”^”hisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãI\J\LÓLÓNJOJQÁQÁS8T8V¯V¯X&Y&[�[�]^`‹`‹bceyeygðgðigjglÞlÞnPoPqÂqÂs4t4v¦v¦xy{Š{Š}ü}ün€n‚à‚à„R…R‡ćĉ6Š6Œ¨Œ¨Ž�‘Œ‘Œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!1×}!1Ð×}!1u¾!1Lu¾!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1©ç!1(©ç!1G("1¤G("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1{Q"1€{Q"1’"1ü’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1aÔ"1Ø aÔ"1ÇÚ"1° ÇÚ"1'ÿ"1, 'ÿ"1e-#1Le-#1+4#1$+4#1_‡#1�_‡#1Õ$1Õ$1­ò$1à­ò$1çÞ%1äçÞ%1Û€&1ÄÛ€&1«R'1«R'1Se'1 €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÝ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @b¢C bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÝ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @b¢C bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1fšÝ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @b¢C  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 /++s�¡ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1®ò€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1¸8¦&”‚^ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛQQÇÇ)=*=4³4³>)?)I�I�ST^…^…hùhùrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹wwííc c"Ù"Ù$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒuŒuŽéŽé�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lIi!11~!1Ð1~!1Ͼ!1LϾ!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1è!1(è!1¡("1¤¡("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1ÕQ"1€ÕQ"1s’"1üs’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1eÓ"1Ø eÓ"1ÇÚ"1´ ÇÚ"1'ÿ"10 'ÿ"1i,#1Pi,#1?5#1,?5#1��#1˜��#1É$1$É$1¡ò$1è¡ò$1ÛÞ%1ìÛÞ%1±€&1̱€&1�R'1�R'1)e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®j6 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡v¶B bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®j6 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡v¶B bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®j6  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡v¶B  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 <\¤Ä„¤ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1É €� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1$¤-­6¶?¿I‰ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Û ÛPPÅÅ):*:4¯4¯>&?&I›I›ST^‚^‚hôhôrfsf} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÛÛPPÅÅ : : ¯ ¯$$™™ƒƒøømmââW W"Ì"Ì$A%A'¶'¶)+*+, , ./1Š1Š3ÿ3ÿ5t6t8é8é:^;^=Ó=Ó?H@HB½B½D2E2G§G§IJL“L“NOQ}Q}SòSòUgVgXÜXÜZQ[Q]Æ]Æ_;`;b°b°d%e%gšgšijl„l„nönöphqhsÚsÚuLvLx¾x¾z0{0}¢}¢€‚†‚†„ø„ø†j‡j‰܉܋NŒNŽÀŽÀ�2‘2“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lOi!1Ý}!1ÐÝ}!1{¾!1L{¾!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1¯ç!1(¯ç!1M("1¤M("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1�Q"1€�Q"1’"1ü’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1eÓ"1Ø eÓ"1ÇÚ"1´ ÇÚ"1'ÿ"18 'ÿ"1i,#1Xi,#1{2#14{2#1‡#1 ‡#1á$1,á$1¹ò$1ð¹ò$1óÞ%1ôóÞ%1�&1Ô�&1çR'1çR'1�e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›à b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @b¢C bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›à b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @b¢C bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›à  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @b¢C  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7MGoperation.3exchange :z2breasonOperatorInputsvalue7invocations1D@@Пé� bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€ùyòrëkÈ €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ü ÜSSÊÊ)A*A4¸4¸>/?/I¤I¤ST^Ž^Žhisxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÜÜSSÊÊ A A ¸ ¸//¦¦””  ‚‚ùùp p"ç"ç$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbébéd^e^gÓgÓiHjHl½l½n2o2q§q§stv‘v‘xy{{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lOi!1%~!1Ð%~!1þ!1Lþ!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1÷ç!1(÷ç!1•("1¤•("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1ÉQ"1€ÉQ"1g’"1üg’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1eÓ"1Ø eÓ"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1{2#1,{2#1w�#1˜w�#1ç$1$ç$1¿ò$1è¿ò$1ùÞ%1ìùÞ%1�&1Ì�&1íR'1íR'1•e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡[' b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡¡áC bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡[' b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡¡áC bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡['  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡¡áC  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ?nŽ‚¢ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1@€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1€”(¨<¼0p €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll æ æ``ÖÖ)L*L4Â4Â>8?8I¬I¬S T ^”^”his|s|} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllææ``ÚÚ T T Î ÎHHÂÂ<<¶¶00ªª$$ ž ž"#%Ž%Ž'(*z*z,ð,ð.f/f1Ü1Ü3R4R6È6È8>9>;´;´=*>*@ @ BCEŒEŒGHJxJxLìLìN`O`QÔQÔSHTHV¼V¼X0Y0[¤[¤]^`Œ`Œbcetetgègèi\j\lÐlÐnDoDq¸q¸s,t,v v xy{ˆ{ˆ}ü}üp€p‚ä‚ä„X…X‡̇̉@Š@Œ´Œ´Ž(�(‘œ‘œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lOi!1Ý}!1ÐÝ}!1{¾!1L{¾!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1¯ç!1(¯ç!1M("1¤M("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1�Q"1€�Q"1’"1ü’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1aÔ"1Ø aÔ"1ÇÚ"1° ÇÚ"1'ÿ"1, 'ÿ"1e-#1Le-#174#1$74#1ˇ#1�ˇ#1Ï$1Ï$1§ò$1à§ò$1áÞ%1äáÞ%1±€&1ı€&1�R'1�R'1)e'1 €} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›Þ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 @b¢C bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›Þ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 @b¢C bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›Þ  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 @b¢C  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÂ)gÁoperation.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€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 $  h€  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÂ)gÁoperation.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€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1¹ý€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1ädÝ]ÖVÏO�Á €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜRRÇÇ)<*<4±4±>&?&I™I™S T ^^hòhòrese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒxxîîd d"Ú"Ú$O%O'Ä'Ä)9*9,®,®.#/#1˜1˜3 4 6‚6‚8÷8÷:l;l=á=á?V@VBËBËD@E@GµGµI*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b¨b¨degŽgŽijltltnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·lOi!1+~!1Ð+~!1ɾ!1Lɾ!1¿Å!1T¿Å!1IÏ!1ÐIÏ!1Ò!1œÒ!1÷ç!1(÷ç!1•("1¤•("1‘/"1¬‘/"19"1(9"1Q<"1ôQ<"1ÉQ"1€ÉQ"1g’"1üg’"1c™"1 c™"1í¢"1€ í¢"1#¦"1L #¦"1eÓ"1Ø eÓ"1ÍÚ"1´ ÍÚ"1-ÿ"10 -ÿ"1o,#1Po,#1Q5#1,Q5#1I‘#1˜I‘#1Ï$1$Ï$1§ò$1è§ò$1áÞ%1ìáÞ%1¥€&1Ì¥€&1uR'1uR'1e'1€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£_+ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 a¡¡áC bÊwµoperation.0ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue4kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.1ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue5bytes2304patternContiguousinvocations1€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£_+ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.1exchange :z2breasonOperatorInputsvalue5invocations1 a¡¡áC bÊwµoperation.1ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue5kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1  bšWuoperation.2ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue6bytes2304patternContiguousinvocations1 €} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1o£_+  b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1 €_ bJ7M7operation.2exchange :z2breasonOperatorInputsvalue6invocations1 a¡¡áC  bÊwµoperation.2ipu_stack_reduce_sum_f165:5z9299Z=2MzQ*QzU*UzY*Yb]reasonOperatorKernelvalue6kernelSpecPlanned(ReductionSum { partials: 2 })outputElements1152input0Elements1152input1Elements1152invocations1 bšWuoperation.3ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue7bytes2304patternContiguousinvocations1 bÊ-gÅoperation.3ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue7kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Ê bJ7M7operation.3exchange :z2breasonOperatorInputsvalue7invocations1 ?_Ž®ƒ£ bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 bÊ-gÅoperation.4ipu_stack_unpack_amp_to_row_major_f16_o1_r16_p16_c72_p72-:-z1211Z5ZYz]*]za*abereasonOperatorKernelvalue8kernelSpecPlanned(Rearrange { from: Layout { order: Amp(TransposedLeft), tiling: TensorTiling { tile_count: 736, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 46, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 16, block_size: 2, padding_multiple: 2, shard_padding_multiple: 1, padding: Zero, tile_stride: Some(46) }] }, memory_class: Ipu21Interleaved }, to: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard } })outputElements1152input0Elements1152invocations1€Q bJ7Moperation.4exchange :z2breasonOperatorInputsvalue8invocations1Ô€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1PÐdäxøŒ D €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj â âZZÎÎ)B*B4¶4¶>*?*IœIœST^€^€hòhòrdsd} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjââZZÒÒ J J  Â::²²**¢¢’’  ‚ ‚"ú"ú$n%n'â'â)V*V,Ê,Ê.>/>1²1²3&4&6š6š89;‚;‚=ö=ö?j@jBÞBÞDRERGÆGÆI:J:L¬L¬NOQ�Q�STVtVtXæXæZX[X]Ê]Ê_<`v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1û}!1�û}!1™¾!1™¾!1¿Â!1$¿Â!1Çç!1�Çç!1e("1e("1‘,"1$‘,"1™Q"1�™Q"17’"17’"1c–"1$c–"1#Ù"1�#Ù"1ƒý"1üƒý"1;0#1 ;0#1#1| #1ó$1Ü ó$1Ëò$1  Ëò$1ß%1¤ ß%1 �&1„  �&1ÛR'1ÈÛR'1ƒe'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14y­&òÕ¡ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14y­&òÕ¡ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14y­&òÕ¡ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10zz ô ôjjàà)V*V4Ê4Ê>>?>I²I²S&T&^š^šhis€s€} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðzzôônnèè b b Ü ÜVVÐÐJJÄÄ::°°&& œ œ"#%ˆ%ˆ'þ'þ)t*t,ê,ê.`/`1Ö1Ö3L4L6Â6Â8898;®;®=">"@–@–B C E~E~GòGòIfJfLÚLÚNNONQÂQÂS6T6VªVªXY[’[’]^`z`zbîbîdbebgÖgÖiJjJl¾l¾n2o2q¦q¦stvŽvŽxy{v{v}ê}ê^€^‚Ò‚Ò„F…F‡¸‡¸‰+Š+ŒžŒžŽ�‘„‘„“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1~!1�~!1½¾!1 ½¾!1MÅ!1MÅ!1ñç!1€ñç!1�("1ü�("1/"1/"1ÃQ"1pÃQ"1a’"1ìa’"1ñ˜"1ôñ˜"1#Ù"1`#Ù"1ƒý"1̃ý"1;0#1ì;0#1#1L #1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1±€&1T ±€&1�R'1˜�R'1)e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•O b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1�Í€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•O b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1�Í€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•O b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1�Í €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr è è]]ÒÒ)G*G4º4º>-?-I I ST^†^†hùhùrjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrèè^^ÔÔ J J À À66¬¬""˜˜  ‚‚÷÷l l"á"á$V%V'Ë'Ë)@*@,µ,µ.*/*1Ÿ1Ÿ346‰6‰8þ8þ:s;s=æ=æ?Y@YBÌBÌD?E?G²G²I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b£b£deg‰g‰iüiükolonânâpUqUsÈsÈu;v;x®x®z!{!}”}”€‚z‚z„í„í†^‡^‰ЉЋBŒBŽ´Ž´�&‘&“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1õ}!1�õ}!1“¾!1“¾!1¿Â!1$¿Â!1Çç!1�Çç!1e("1e("1‘,"1$‘,"1™Q"1�™Q"17’"17’"1c–"1$c–"1#Ù"1�#Ù"1ƒý"1üƒý"1;0#1 ;0#1#1| #1Û$1Ü Û$1³ò$1  ³ò$1íÞ%1¤ íÞ%1“€&1„ “€&1cR'1ÈcR'1 e'1Ì€} bJ7UGoperation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14w«"îÖ¢ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7UGoperation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14w«"îÖ¢ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7UGoperation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations14w«"îÖ¢ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10xx ð ðddØØ)L*L4¾4¾>0?0I¢I¢ST^†^†høhørhsh} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðxxððhhàà X X Ð ÐHHÀÀ88°°$$˜˜  € €"ô"ô$h%h'Ü'Ü)P*P,Ä,Ä.8/81¬1¬3 4 6”6”89;|;|=î=î?`@`BÒBÒDDEDG¶G¶I(J(LšLšN O Q~Q~SðSðUbVbXÔXÔZF[F]¸]¸_*`*bœbœdeg€g€iòiòkdldnÖnÖpHqHsºsºu,v,xžxžz{}‚}‚ôô�f‚f„؄؆H‡H‰¹‰¹‹*Œ*ޛޛ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1%~!1�%~!1þ!1 þ!1MÅ!1MÅ!1ñç!1€ñç!1�("1ü�("1/"1/"1ÃQ"1pÃQ"1a’"1ìa’"1ñ˜"1ôñ˜"1#Ù"1`#Ù"1ƒý"1̃ý"1;0#1ì;0#1#1L #1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1K€&1T K€&1R'1˜R'1Ãd'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Q b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ŒÌ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Q b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ŒÌ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1a•Q b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ŒÌ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ê ê]]ÐÐ)C*C4´4´>%?%I–I–ST^x^xhéhérXsX} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðttêê``ÖÖ L L  Â88®®$$šš  €€óóf f"Ù"Ù$L%L'¿'¿)2*2,¥,¥./1‹1‹3þ3þ5q6q8ä8ä:W;W=È=È?9@9BªBªDEGŒGŒIýIýKnLnNßNßPPQPSÁSÁU2V2X£X£Z[]…]…_ö_öagbgdØdØfIgIiºiºk+l+nœnœp q s~s~uïuïw`x`zÑzÑ|B}B³³�$‚$„•„•†‡‰t‰t‹ä‹ä�TŽT�Ä�Ä’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1ï}!1�ï}!1�¾!1 �¾!1¿Â!1¿Â!1Áç!1€Áç!1_("1ü_("1‘,"1‘,"1“Q"1p“Q"11’"1ì1’"1c–"1ôc–"1Ù"1`Ù"1}ý"1Ì}ý"1;0#1ì;0#1#1L #1ù$1¬ ù$1Ñò$1p Ñò$1 ß%1t  ß%1S�&1T S�&1#S'1˜#S'1Ëe'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vª#ï b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vª#ï b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vª#ï b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10vv ï ïhháá)Z*Z4Ó4Ó>J?JIÁIÁS6T6^©^©his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðvvïïhháá Z Z Ó ÓLLÅÅ>>··00©©"" › ›"#%�%�'(**,ø,ø.q/q1ê1ê3c4c6Ü6Ü8U9U;Î;Î=G>G@À@ÀB9C9E²E²G+H+J¢J¢LMO�O�QRT~T~VõVõXlYl[ã[ã]Z^Z`Ñ`ÑbFcFe»e»g0h0j¥j¥lmo�o�qrtstsvævæxYyY{Ì{Ì}?~?€²€²‚%ƒ%…˜…˜‡ ˆ Š{Š{ŒíŒíŽ_�_‘ёѓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1%~!1�%~!1þ!1 þ!1_Å!1_Å!1÷ç!1€÷ç!1•("1ü•("11/"11/"1ÉQ"1pÉQ"1g’"1ìg’"1™"1ô™"1Ù"1`Ù"1}ý"1Ì}ý"1;0#1ì;0#1#1L #1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1í€&1T í€&1½R'1˜½R'1ee'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžX$ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¡á€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžX$ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¡á€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžX$ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¡á €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq é é^^ÓÓ)H*H4½4½>1?1I¥I¥ST^�^�hisusu} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqééaaÙÙ Q Q É ÉAA¹¹11©©““ } }"ò"ò$g%g'Ü'Ü)Q*Q,Æ,Æ.;/;1°1°3%4%6š6š89;„;„=ù=ù?n@nBãBãDXEXGÍGÍIAJALµLµN)O)Q�Q�STV…V…XùXùZm[m]á]á_U`UbÉbÉd=e=g±g±i%j%l™l™n o q�q�sõsõuivixÝxÝzQ{Q}Å}Å9€9‚­‚­„!…!‡•‡•‰ Š ŒŒŽôŽô�i‘i“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1ï}!1�ï}!1�¾!1 �¾!1¿Â!1¿Â!1»ç!1€»ç!1Y("1üY("1‘,"1‘,"1�Q"1p�Q"1+’"1ì+’"1c–"1ôc–"1Ù"1`Ù"1}ý"1Ì}ý"1;0#1ì;0#1#1L #1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1á€&1T á€&1±R'1˜±R'1Ye'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨ë b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨ë b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨ë b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10tt ë ëbbÙÙ)P*P4Ç4Ç>>µµ,,££‘‘  "ö"ö$m%m'ä'ä)[*[,Ò,Ò.I/I1À1À37476®6®8%9%;œ;œ=>@Š@ŠBCExExGïGïIdJdLÙLÙNNONQÃQÃS8T8V­V­X"Y"[—[—] ^ `�`�bôbôdgeggÚgÚiMjMlÀlÀn1o1q¢q¢stv„v„xõxõzf{f}×}×H€H‚¹‚¹„*…*‡š‡š‰ Š Œ|Œ|ŽíŽí�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LUi!1+~!1�+~!1ɾ!1 ɾ!1_Å!1_Å!1÷ç!1€÷ç!1•("1ü•("11/"11/"1ÉQ"1pÉQ"1g’"1ìg’"1™"1ô™"1Ù"1`Ù"1}ý"1Ì}ý"1;0#1ì;0#1#1L #1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1«€&1T «€&1{R'1˜{R'1#e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžZ& b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¡á€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžZ& b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¡á€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžZ& b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¡á €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10qq ç ç\\ÑÑ)F*F4»4»>.?.I¡I¡ST^…^…h÷h÷risi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðqqçç]]ÓÓ I I ¿ ¿55««!!——  ��öök k"à"à$U%U'Ê'Ê)?*?,´,´.)/)1ž1ž346ˆ6ˆ8ý8ý:r;r=ç=ç?\@\BÑBÑDFEFG»G»I.J.L¡L¡NOQ‡Q‡SúSúUmVmXàXàZS[S]Æ]Æ_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰̉̋?Œ?޲޲�%‘%“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1é}!1�é}!1‡¾!1 ‡¾!1ÅÂ!1ÅÂ!1»ç!1€»ç!1Y("1üY("1—,"1—,"1�Q"1p�Q"1+’"1ì+’"1i–"1ôi–"1Ù"1`Ù"1wý"1Ìwý"150#1ì50#1#1L #1ù$1¬ ù$1Ñò$1p Ñò$1 ß%1t  ß%1_�&1T _�&1/S'1˜/S'1×e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§ ì b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 J€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§ ì b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 J€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§ ì b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 J €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10rr ê êbbÚÚ)R*R4Ê4Ê>@?@I¶I¶S,T,^¢^¢hisŽsŽ} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðrrêêbbÚÚ R R Ê ÊBBºº22ªª""šš Š Š"#%z%z'ò'ò)j*j,â,â.Z/Z1Ò1Ò3J4J6Â6Â8:9:;²;²=*>*@¢@¢BCE’E’G H J€J€LöLöNlOlQâQâSXTXVÎVÎXDYD[º[º]0^0`¦`¦bce’e’ghj~j~lôlônjojqàqàsVtVvÌvÌxByB{¸{¸}.~.€¤€¤‚ƒ…�…�‡ˆŠ|Š|ŒòŒòŽh�h‘Þ‘Þ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1+~!1�+~!1ɾ!1 ɾ!1/Å!1/Å!1ýç!1€ýç!1›("1ü›("1/"1/"1ÏQ"1pÏQ"1m’"1ìm’"1Ó˜"1ôÓ˜"1Ù"1`Ù"1wý"1Ìwý"150#1ì50#1#1L #1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1�&1T �&1áR'1˜áR'1‰e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§a- b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1uµ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§a- b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1uµ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§a- b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1uµ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ä ä[[ÒÒ)I*I4À4À>7?7I®I®S%T%^—^—h i s{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmää[[ÒÒ I I À À77®®%%œœŠŠ x x"ï"ï$f%f'Ý'Ý)T*T,Ë,Ë.B/B1¹1¹30406§6§89;•;•= > @ƒ@ƒBúBúDqEqGèGèI_J_LÖLÖNMOMQÄQÄS;T;V²V²X)Y)[ [ ]^`Ž`Žbce|e|gógóijjjlálánSoSqÅqÅs7t7v©v©xy{�{�}ÿ}ÿq€q‚ã‚ã„U…U‡LJlj9Š9Œ«Œ«Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1é}!1�é}!1‡¾!1 ‡¾!1ÅÂ!1ÅÂ!1µç!1€µç!1S("1üS("1—,"1—,"1‡Q"1p‡Q"1%’"1ì%’"1i–"1ôi–"1Ù"1`Ù"1wý"1Ìwý"150#1ì50#1#1L #1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1ç€&1T ç€&1·R'1˜·R'1_e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥è b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 K€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥è b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 K€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1q¥è b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 K €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10pp æ æ\\ÒÒ)H*H4¾4¾>2?2I¦I¦ST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðppææ\\ÒÒ H H ¾ ¾44ªª  ––  ‚‚øøn n"ä"ä$Z%Z'Ð'Ð)F*F,¼,¼.2/21¨1¨346”6”8 9 ;€;€=ö=ö?l@lBâBâDXEXGÎGÎIBJBL¶L¶N*O*QžQžSTV†V†XúXúZn[n]â]â_V`VbÊbÊd>e>g²g²i&j&lšlšnoq‚q‚sösöujvjxÞxÞzR{R}Æ}Æ:€:‚®‚®„"…"‡–‡–‰ Š Œ~Œ~ŽòŽò�f‘f“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LOi!1+~!1�+~!1ɾ!1 ɾ!1/Å!1/Å!1ýç!1€ýç!1›("1ü›("1/"1/"1ÏQ"1pÏQ"1m’"1ìm’"1Ó˜"1ôÓ˜"1Ù"1`Ù"1wý"1Ìwý"150#1ì50#1#1L #1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1·€&1T ·€&1‡R'1˜‡R'1/e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§c/ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1uµ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§c/ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1uµ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1s§c/ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1uµ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm â âWWÌÌ)A*A4¶4¶>+?+I I ST^‡^‡hùhùrksk} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmââWWÌÌ A A ¶ ¶++  ŠŠÿÿttéé^ ^"Ó"Ó$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?O@OBÄBÄD9E9G®G®I#J#L˜L˜N O Q‚Q‚S÷S÷UlVlXáXáZV[V]Ë]Ë_@`@bµbµd*e*gŸgŸijl‰l‰nûnûpmqmsßsßuQvQxÃxÃz5{5}§}§€‚‹‚‹„ý„ý†o‡o‰á‰á‹SŒSŽÅŽÅ�7‘7“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1ã}!1�ã}!1�¾!1 �¾!1³Â!1³Â!1µç!1€µç!1S("1üS("1…,"1…,"1‡Q"1p‡Q"1%’"1ì%’"1W–"1ôW–"1Ù"1`Ù"1wý"1Ìwý"1/0#1ì/0#1û~#1L û~#1ç$1¬ ç$1¿ò$1p ¿ò$1ùÞ%1t ùÞ%1�&1T �&1íR'1˜íR'1•e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤é b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤é b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1p¤é b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn å å\\ÓÓ)J*J4Á4Á>6?6I«I«S T ^•^•h i ss} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnåå\\ÓÓ J J Á Á88¯¯&&��‹‹ y y"ð"ð$g%g'Þ'Þ)U*U,Ì,Ì.C/C1º1º31416¨6¨89;–;–= > @„@„BûBûDrErGéGéI^J^LÓLÓNHOHQ½Q½S2T2V§V§XY[‘[‘]^`{`{bðbðdeeegÚgÚiOjOlÄlÄn9o9q®q®s#t#v˜v˜x y {‚{‚}÷}÷l€l‚á‚á„V…V‡ˇˉ@Š@ŒµŒµŽ*�*‘Ÿ‘Ÿ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!11~!1�1~!1Ͼ!1 Ͼ!1_Å!1_Å!1è!1€è!1¡("1ü¡("11/"11/"1ÕQ"1pÕQ"1s’"1ìs’"1™"1ô™"1Ù"1`Ù"1wý"1Ìwý"1/0#1ì/0#1û~#1L û~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1�&1T �&1çR'1˜çR'1�e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°j6 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1ªê€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°j6 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1ªê€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°j6 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1ªê €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ç çaa××)M*M4Ã4Ã>7?7I«I«ST^“^“his{s{} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmççaaÛÛ U U Ï ÏIIÃÃ==··11««%% Ÿ Ÿ"#%�%�'(*{*{,ñ,ñ.g/g1Ý1Ý3S4S6É6É8?9?;µ;µ=+>+@¡@¡BCE�E�GHJwJwLëLëN_O_QÓQÓSGTGV»V»X/Y/[£[£]^`‹`‹bÿbÿdsesgçgçi[j[lÏlÏnCoCq·q·s+t+vŸvŸxy{‡{‡}û}ûo€o‚ã‚ã„W…W‡ˇˉ?Š?Œ³Œ³Ž'�'‘›‘›“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ý}!1�Ý}!1{¾!1 {¾!1³Â!1³Â!1¯ç!1€¯ç!1M("1üM("1…,"1…,"1�Q"1p�Q"1’"1ì’"1W–"1ôW–"1Ù"1`Ù"1wý"1Ìwý"1/0#1ì/0#1û~#1L û~#1Ï$1¬ Ï$1§ò$1p §ò$1áÞ%1t áÞ%1±€&1T ±€&1�R'1˜�R'1)e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢å b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢å b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢å b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm ã ãYYÎÎ)C*C4¸4¸>+?+IžIžST^„^„h÷h÷rjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmããYYÏÏ E E » »11§§““  õõk k"á"á$V%V'Ë'Ë)@*@,µ,µ.*/*1Ÿ1Ÿ346‰6‰8þ8þ:s;s=è=è?]@]BÒBÒDGEGG¼G¼I/J/L¢L¢NOQˆQˆSûSûUnVnXáXáZT[T]Ç]Ç_:`:b­b­d e g“g“ijlylynìnìp_q_sÒsÒuEvEx¸x¸z+{+}ž}ž€‚„‚„„÷„÷†j‡j‰݉݋PŒPŽÃŽÃ�6‘6“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!11~!1�1~!1Ͼ!1 Ͼ!1_Å!1_Å!1è!1€è!1¡("1ü¡("11/"11/"1ÕQ"1pÕQ"1s’"1ìs’"1™"1ô™"1Ù"1`Ù"1wý"1Ìwý"1/0#1ì/0#1û~#1L û~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Ÿ€&1T Ÿ€&1oR'1˜oR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°l8 b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1«ë€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°l8 b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1«ë€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1|°l8 b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1«ë €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Ññ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10mm å å]]ÑÑ)E*E4¹4¹>+?+I�I�ST^�^�hóhórese} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðmmåå]]ÕÕ M M Å Å==µµ--¥¥••  … …"ý"ý$q%q'å'å)Y*Y,Í,Í.A/A1µ1µ3)4)6�6�89;…;…=ù=ù?m@mBáBáDUEUGÉGÉI;J;L­L­NOQ‘Q‘STVuVuXçXçZY[Y]Ë]Ë_=`=b¯b¯d!e!g“g“ijlwlwnénép[q[sÍsÍu?v?x±x±z#{#}•}•€‚y‚y„ë„ë†]‡]‰ωϋAŒA޳޳�%‘%“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1Ý}!1�Ý}!1{¾!1 {¾!1ÅÂ!1ÅÂ!1¯ç!1€¯ç!1M("1üM("1—,"1—,"1�Q"1p�Q"1’"1ì’"1i–"1ôi–"1£Ú"1`£Ú"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1ó€&1T ó€&1ÃR'1˜ÃR'1ke'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡æ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1Q€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡æ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1Q€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1m¡æ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1Q €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1cƒ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10nn è è^^ÔÔ)J*J4¾4¾>2?2I¦I¦ST^Ž^Žhistst} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðnnèèbbÜÜ V V Ð ÐJJÄÄ>>¸¸..¤¤ � �"#%|%|'ò'ò)h*h,Þ,Þ.T/T1Ê1Ê3@4@6¶6¶8,9,;¢;¢=>@Š@ŠBþBþDrErGæGæIZJZLÎLÎNBOBQ¶Q¶S*T*VžVžXY[†[†]ú]ú_n`nbâbâdVeVgÊgÊi>j>l²l²n&o&qšqšstv‚v‚xöxözj{j}Þ}ÞR€R‚ƂƄ:…:‡¬‡¬‰ŠŒ’Œ’Ž�‘x‘x“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!17~!1�7~!1Õ¾!1 Õ¾!1)Å!1)Å!1è!1€è!1¡("1ü¡("1û."1û."1ÕQ"1pÕQ"1s’"1ìs’"1͘"1ô͘"1�Ú"1`�Ú"1ýþ"1Ìýþ"1/0#1ì/0#1û~#1L û~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1“€&1T “€&1cR'1˜cR'1 e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹s? b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1v¶€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹s? b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1v¶€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1…¹s? b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1v¶ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1\|  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Ü ÜQQÆÆ);*;4®4®>!?!I”I”ST^z^zhíhír^s^} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÜÜRRÈÈ > > ´ ´**  ŒŒvvëë` `"Õ"Õ$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ú=Ú?M@MBÀBÀD3E3G¦G¦IJLŒLŒNÿNÿPrQrSåSåUXVXXËXËZ>[>]±]±_$`$b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†R‡R‰ĉċ6Œ6ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LIi!1×}!1�×}!1u¾!1 u¾!1¿Â!1¿Â!1©ç!1€©ç!1G("1üG("1‘,"1‘,"1{Q"1p{Q"1’"1ì’"1c–"1ôc–"1µÚ"1`µÚ"1ÿ"1Ìÿ"1/0#1ì/0#1û~#1L û~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1{€&1T {€&1KR'1˜KR'1ód'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸâ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1P€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸâ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1P€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1kŸâ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1P €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1t”  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll ä äXXÌÌ)@*@4²4²>$?$I–I–ST^z^zhìhìr\s\} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllää\\ÔÔ L L Ä Ä<<´´,,¤¤ŒŒ t t"è"è$\%\'Ð'Ð)D*D,¸,¸.,/,1 1 346ˆ6ˆ8ü8ü:p;p=â=â?T@TBÆBÆD8E8GªGªIJLŽLŽNOQrQrSäSäUVVVXÈXÈZ:[:]¬]¬_`b�b�degtgtiæiækXlXnÊnÊp?IŠIŠSûSû]l^lhÝhÝrLsL} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhÞÞTTÊÊ @ @ ¶ ¶,,¢¢ŽŽttççZ Z"Í"Í$@%@'³'³)&*&,™,™. / 113ò3ò5e6e8Ø8Ø:K;K=¼=¼?-@-BžBžDEG€G€IñIñKbLbNÓNÓPDQDSµSµU&V&X—X—Z[]y]y_ê_êa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1×}!1�×}!1u¾!1 u¾!1­Â!1­Â!1£ç!1€£ç!1A("1üA("1,"1,"1uQ"1puQ"1’"1ì’"1Q–"1ôQ–"1£Ú"1`£Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1á$1¬ á$1¹ò$1p ¹ò$1óÞ%1t óÞ%1;�&1T ;�&1 S'1˜ S'1³e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžã b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžã b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1jžã b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1hˆ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10jj ã ã\\ÕÕ)N*N4Ç4Ç>>?>IµIµS*T*^�^�his�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðjjãã\\ÕÕ N N Ç Ç@@¹¹22««$$�� � �"#%�%�'ú'ú)s*s,ì,ì.e/e1Þ1Þ3W4W6Ð6Ð8I9I;Â;Â=;>;@´@´B-C-E¦E¦GHJ–J–L M O„O„QûQûSrTrVéVéX`Y`[×[×]N^N`Å`Åb:c:e¯e¯g$h$j™j™lmo�o�qôqôsgtgvÚvÚxMyM{À{À}3~3€¦€¦‚ƒ…Œ…Œ‡ý‡ý‰oŠoŒáŒáŽS�S‘őœ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1=~!1�=~!1Û¾!1 Û¾!1SÅ!1SÅ!1 è!1€ è!1§("1ü§("1%/"1%/"1ÛQ"1pÛQ"1y’"1ìy’"1÷˜"1ô÷˜"1�Ú"1`�Ú"1ýþ"1Ìýþ"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Ï€&1T Ï€&1ŸR'1˜ŸR'1Ge'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂ|H b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¦æ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂ|H b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¦æ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂ|H b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¦æ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1a�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Ý ÝRRÇÇ)<*<4±4±>%?%I™I™S T ^�^�hõhõrisi} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÝÝUUÍÍ E E ½ ½55­­%%��‡‡üüq q"æ"æ$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`Ib½b½d1e1g¥g¥ijl�l�noququséséu]v]xÑxÑzE{E}¹}¹-€-‚¡‚¡„…‡‰‡‰‰þ‰þ‹sŒsŽèŽè�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1Ñ}!1�Ñ}!1o¾!1 o¾!1­Â!1­Â!1£ç!1€£ç!1A("1üA("1,"1,"1uQ"1puQ"1’"1ì’"1Q–"1ôQ–"1»Ú"1`»Ú"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1É€&1T É€&1™R'1˜™R'1Ae'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœß b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœß b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1hœß b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1�¡  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10hh ß ßVVÍÍ)D*D4»4»>0?0I¥I¥ST^‰^‰húhúrjsj} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðhhßßVVÍÍ D D » »22©©  ——……üüs s"ê"ê$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n%o%q–q–stvxvxxéxézZ{Z}Ë}Ë<€<‚­‚­„…‡އމÿ‰ÿ‹pŒpŽáŽá�R‘R“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LCi!1=~!1�=~!1Û¾!1 Û¾!1SÅ!1SÅ!1è!1€è!1­("1ü­("1%/"1%/"1áQ"1páQ"1’"1ì’"1÷˜"1ô÷˜"1µÚ"1`µÚ"1ÿ"1Ìÿ"1)0#1ì)0#1õ~#1L õ~#1½$1¬ ½$1•ò$1p •ò$1ÏÞ%1t ÏÞ%1�€&1T �€&1]R'1˜]R'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂ~J b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1§ç€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂ~J b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1§ç€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1ŽÂ~J b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1§ç €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1zš  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ee Û ÛPPÅÅ):*:4¯4¯>"?"I•I•ST^y^yhëhër]s]} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðeeÛÛQQÇÇ = = ³ ³))ŸŸ‹‹uuêê_ _"Ô"Ô$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?P@PBÅBÅD:E:G¯G¯I"J"L•L•NOQ{Q{SîSîUaVaXÔXÔZG[G]º]º_-`-bŸbŸdegƒgƒiõiõkglgnÙnÙpKqKs½s½u/v/x¡x¡z{}…}…÷÷�i‚i„Û„Û†M‡M‰À‰À‹3Œ3ަަ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Ñ}!1�Ñ}!1o¾!1 o¾!1¹Â!1¹Â!1�ç!1€�ç!1;("1ü;("1‹,"1‹,"1oQ"1poQ"1 ’"1ì ’"1]–"1ô]–"1©Ú"1`©Ú"1 ÿ"1Ì ÿ"1#0#1ì#0#1ï~#1L ï~#1Û$1¬ Û$1³ò$1p ³ò$1íÞ%1t íÞ%1G�&1T G�&1S'1˜S'1¿e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›à b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1N€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›à b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1N€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1g›à b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1N €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1o�  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff Þ ÞVVÎÎ)F*F4¾4¾>4?4IªIªS T ^˜^˜his„s„} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffÞÞVVÎÎ F F ¾ ¾66®®&&žžŽŽ ~ ~"ö"ö$n%n'æ'æ)^*^,Ö,Ö.N/N1Æ1Æ3>4>6¶6¶8.9.;¦;¦=>@–@–BCE†E†GþGþItJtLêLêN`O`QÖQÖSLTLVÂVÂX8Y8[®[®]$^$`š`šbce†e†gügüirjrlèlèn`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1=~!1�=~!1Û¾!1 Û¾!1Å!1Å!1è!1€è!1­("1ü­("1ï."1ï."1áQ"1páQ"1’"1ì’"1Á˜"1ôÁ˜"1£Ú"1`£Ú"1ÿ"1Ìÿ"1#0#1ì#0#1ï~#1L ï~#1Õ$1¬ Õ$1­ò$1p ­ò$1çÞ%1t çÞ%1ÿ€&1T ÿ€&1ÏR'1˜ÏR'1we'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—Ë…Q b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1uµ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—Ë…Q b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1uµ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—Ë…Q b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1uµ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1hˆ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ø ØOOÆÆ)=*=4´4´>+?+I¢I¢ST^•^•hisysy} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaØØOOÆÆ = = ´ ´++¢¢��~~õõl l"ã"ã$Z%Z'Ñ'Ñ)H*H,¿,¿.6/61­1­3$4$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜISJSLÊLÊNAOAQ¸Q¸S/T/V¦V¦XY[”[”] ^ `‚`‚bùbùdpepgçgçi^j^lÕlÕnQoQqÃqÃs5t5v§v§xy{‹{‹}ý}ýo€o‚á‚á„S…S‡Ňʼn7Š7Œ©Œ©Ž�‘�‘�“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Ë}!1�Ë}!1i¾!1 i¾!1¹Â!1¹Â!1�ç!1€�ç!1;("1ü;("1‹,"1‹,"1oQ"1poQ"1 ’"1ì ’"1]–"1ô]–"1ÇÚ"1`ÇÚ"1'ÿ"1Ì'ÿ"1#0#1ì#0#1ï~#1L ï~#1É$1¬ É$1¡ò$1p ¡ò$1ÛÞ%1t ÛÞ%1Ï€&1T Ï€&1ŸR'1˜ŸR'1Ge'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ü b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1 M€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ü b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1 M€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1e™Ü b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1 M €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1Œ¬  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10dd Ú ÚPPÆÆ)<*<4²4²>&?&IšIšST^†^†húhúrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðddÚÚPPÆÆ < < ² ²((žžŠŠvvììb b"Ø"Ø$N%N'Ä'Ä):*:,°,°.&/&1œ1œ346ˆ6ˆ8þ8þ:t;t=ê=ê?`@`BÖBÖDLELGÂGÂI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1C~!1�C~!1á¾!1 á¾!1Å!1Å!1è!1€è!1­("1ü­("1ï."1ï."1áQ"1páQ"1’"1ì’"1Á˜"1ôÁ˜"1»Ú"1`»Ú"1ÿ"1Ìÿ"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1›ò$1p ›ò$1ÕÞ%1t ÕÞ%1¥€&1T ¥€&1uR'1˜uR'1e'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—ˇS b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1uµ€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—ˇS b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1uµ€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1—ˇS b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1uµ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1…¥  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r8_p8_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, AxisTiling { axis: FromEnd(1), partitions: 1, block_size: 16, padding_multiple: 16, shard_padding_multiple: 1, padding: Zero, tile_stride: None }] }, memory_class: Ipu21Standard } })outputElements576input0Elements576invocations1 € b*7Ioperation.4sync :‚2breasonExchangeBarriervalue8invocations1 € b*7Ioperation.5sync :‚2breasonExchangeBarriervalue9invocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ö ÖKKÀÀ)5*54ª4ª>?I”I”S T ^ƒ^ƒhõhõrgsg} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÖÖKKÀÀ 5 5 ª ª””  ~~óóhhÝÝR R"Ç"Ç$<%<'±'±)&*&,›,›./1…1…3ú3ú5o6o8ä8ä:Y;Y=Î=Î?C@CB¸B¸D-E-G¢G¢IJLŒLŒNOQvQvSëSëU`V`XÕXÕZJ[J]¿]¿_4`4b©b©deg“g“ijl}l}n÷n÷piqisÛsÛuMvMx¿x¿z1{1}£}£€‚‡‚‡„ù„ù†k‡k‰݉݋OŒOŽÁŽÁ�3‘3“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L=i!1Å}!1�Å}!1c¾!1 c¾!1§Â!1§Â!1—ç!1€—ç!15("1ü5("1y,"1y,"1iQ"1piQ"1’"1ì’"1K–"1ôK–"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1Õ$1¬ Õ$1ýð$1p ýð$1ÛÀ%1t ÛÀ%1 �&1T  �&1Q'1˜Q'1©c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Ý b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Ý b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1d˜Ý b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ÑÍ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10bb Ù ÙPPÉÉ)@*@4·4·>,?,I¡I¡ST^—^—h i s�s�} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðbbÙÙPPÇÇ > > µ µ,,££‘‘ööm m"ä"ä$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbæbæd[e[gÐgÐiEjElºlºn;o;q°q°s%t%všvšxy{„{„}ù}ùn€n‚ã‚ã„X…X‡͇͉BŠBŒ·Œ·Ž,�,‘¡‘¡“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L=i!1C~!1�C~!1á¾!1 á¾!1GÅ!1GÅ!1è!1€è!1³("1ü³("1/"1/"1çQ"1pçQ"1…’"1ì…’"1ë˜"1ôë˜"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1Û$1¬ Û$1ñ$1p ñ$1áÀ%1t áÀ%1�&1T �&1Q'1˜Q'1¯c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 ÔŽZ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¤ä€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 ÔŽZ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¤ä€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 ÔŽZ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¤ä €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ÑÍ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Û ÛUUÕÕ)K*K4Á4Á>5?5I©I©ST^�^�his…s…} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÛÛUUÏÏ I I à Ã==··11««%%ŸŸ “ “" # %�%�'(*y*y,ï,ï.e/e1Û1Û3Q4Q6Ç6Ç8=9=;³;³=)>)@Ÿ@ŸBCE‹E‹GHJuJuLéLéN]O]QÑQÑSETEV¹V¹X-Y-[¡[¡]^`‰`‰býbýdqeqgågåiYjYlÍlÍnMoMqÁqÁs5t5v©v©xy{‘{‘}~€y€y‚í‚í„a…a‡Õ‡Õ‰IŠIŒ½Œ½Ž1�1‘¥‘¥“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L=i!1Å}!1�Å}!1c¾!1 c¾!1§Â!1§Â!1‘ç!1€‘ç!1/("1ü/("1y,"1y,"1cQ"1pcQ"1’"1ì’"1K–"1ôK–"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1ëð$1p ëð$1ÉÀ%1t ÉÀ%1¥€&1T ¥€&1›P'1˜›P'1Cc'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– Ù b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1@€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– Ù b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1@€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1b– Ù b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1@ €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ÑÍ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa × ×MMÆÆ);*;4°4°>#?#I–I–S T ^ˆ^ˆhûhûrnsn} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaa××MMÃà 9 9 ¯ ¯%%››‡‡ýýsséé_ _"Õ"Õ$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnðnðpcqcsÖsÖuIvIx¼x¼z/{/}¢}¢€‚ˆ‚ˆ„û„û†n‡n‰á‰á‹TŒTŽÇŽÇ�:‘:“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations17L=i!1I~!1�I~!1ç¾!1 ç¾!1MÅ!1MÅ!1è!1€è!1³("1ü³("1/"1/"1çQ"1pçQ"1…’"1ì…’"1ñ˜"1ôñ˜"1áØ"1`áØ"1Aý"1ÌAý"1#0#1ì#0#1ï~#1L ï~#1Ã$1¬ Ã$1ëð$1p ëð$1ÉÀ%1t ÉÀ%1™€&1T ™€&1�P'1˜�P'17c'1œ€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ô�\ b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.0exchange :z2breasonOperatorInputsvalue4invocations1¥å€} bJ7U'operation.1exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ô�\ b-wñoperation.1ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue5kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.1exchange :z2breasonOperatorInputsvalue5invocations1¥å€} bJ7U'operation.2exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1 Ô�\ b-wñoperation.2ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue6kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7Moperation.2exchange :z2breasonOperatorInputsvalue6invocations1¥å €Ê bJ7Moperation.3exchange :z2breasonOperatorInputsvalue7invocations1ÑÍ  bÒ-g operation.3ipu_stack_rearrange_row_major_to_amp_f16_o0_r7_p7_c72_p80-:-z1211Z5¢�z¡"¡z¥"¥b©reasonOperatorInputsvalue7kernelSpecPlanned(Rearrange { from: Layout { order: RowMajor, tiling: TensorTiling { tile_count: 1, replicas: 1, axes: [] }, memory_class: Ipu21Standard }, to: Layout { order: Amp(Left), tiling: TensorTiling { tile_count: 1472, replicas: 1, axes: [AxisTiling { axis: FromEnd(2), partitions: 92, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(16) }, AxisTiling { axis: FromEnd(3), partitions: 16, block_size: 1, padding_multiple: 1, shard_padding_multiple: 1, padding: Reject, tile_stride: Some(1) }, 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 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10aa Ù ÙQQÍÍ)A*A4µ4µ>'?'I™I™S T ^‰^‰hûhûrmsm} b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements5376input0Elements504input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_small_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements5488input0Elements5376invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðaaÙÙQQÉÉ A A ¹ ¹11©©!!™™‰‰ y y"ñ"ñ$m%m'á'á)U*U,É,É.=/=1±1±3%4%6™6™8 9 ;�;�=õ=õ?i@iBÝBÝDQEQGÅGÅI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnñnñpcqcsÕsÕuGvGx¹x¹z+{+}�}�€‚�‚�„ó„ó†e‡e‰׉׋IŒI޻޻�-‘-“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1